training/Hebrew_AC_voice_activation_(public_version).ipynb

{"cells":[{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":4598,"status":"ok","timestamp":1680488252536,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"JquEIDDTKGAR","outputId":"e2368b63-053a-4f04-fdc8-0feac1f5e047"},"outputs":[{"output_type":"stream","name":"stdout","text":["cuda:0\n","cpu\n"]}],"source":["#1\n","import torch\n","import torch.nn as nn\n","import torch.nn.functional as F\n","import numpy as np\n","\n","import matplotlib.pyplot as plt\n","import seaborn as sns\n","import pickle\n","import collections\n","import math\n","import os\n","import time\n","\n","%matplotlib inline\n","\n","# References to the cpu and the running device (a gpu, if one is available)\n","device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n","cpu = torch.device(\"cpu\")\n","print(device)\n","print(cpu)"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":36287,"status":"ok","timestamp":1680488292509,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"mciXO--jR8Cj","outputId":"f1efb8f3-d9b1-43c9-9b7b-20d7b60b3374"},"outputs":[{"output_type":"stream","name":"stdout","text":["Mounted at /root/gdrive\n","YonaVox\n"]}],"source":["#2. mount gdrive\n","# The path references the project folder as if it were under my own directory structure.\n","# In order to access files (a few hundred megabytes) you will have to copy the YonaVox directory to a location of your choice\n","# and update the path accordingly.\n","from google.colab import drive\n","drive.mount('/root/gdrive')\n","#first time copy files from original project: !cp -r ./ColabData/YonaVox/ac_audio/*  ./ColabData/YonaVox/ac_audio1/\n","#in other google account should create shortcut to shared folder: /root/gdrive/.shortcut-targets-by-id/1dBbP3yvAtULEFTCfdX-ZlyuA19YbCpWp\n","!ln -s /root/gdrive/MyDrive/ColabData /content/\n","colab_data = \"./ColabData/YonaVox/\"\n","!ls /content/ColabData"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":7005,"status":"ok","timestamp":1680488305844,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"Frds-2epR7s8","outputId":"400cc61a-f635-4399-c256-a543f3094622"},"outputs":[{"output_type":"stream","name":"stdout","text":["2636\n","294\n"]}],"source":["#3. Load the speech data, created in https://colab.research.google.com/drive/1pxtQLo6CsvQYM5vXkL5g909l_QwyU6mK\n","speech_file = open(colab_data + \"hebrew_speech_train\", \"rb\")\n","tagged_train_spectrograms = np.random.permutation(pickle.load(speech_file))\n","speech_file.close()\n","\n","speech_file = open(colab_data + \"hebrew_speech_test\", \"rb\")\n","tagged_test_spectrograms = np.random.permutation(pickle.load(speech_file))\n","speech_file.close()\n","\n","print(len(tagged_train_spectrograms))\n","print(len(tagged_test_spectrograms))"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/","height":1000},"executionInfo":{"elapsed":2322,"status":"ok","timestamp":1680488314674,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"IZuhgOFx6ZHS","outputId":"bcb6bd58-88df-47f9-98ed-ccb6fcaf3f40"},"outputs":[{"output_type":"stream","name":"stderr","text":["<ipython-input-4-6bd6b97ed52d>:13: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at ../torch/csrc/utils/tensor_new.cpp:230.)\n","  train_spectrograms = normalize_and_permute(torch.tensor(spectrogram_list, dtype=torch.float32))\n"]},{"output_type":"stream","name":"stdout","text":["torch.Size([138, 80])\n","NOISE\n"]},{"output_type":"display_data","data":{"text/plain":["<Figure size 1200x1200 with 1 Axes>"],"image/png":"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\n"},"metadata":{}},{"output_type":"stream","name":"stdout","text":["torch.Size([138, 80])\n","NOISE\n"]},{"output_type":"display_data","data":{"text/plain":["<Figure size 1200x1200 with 1 Axes>"],"image/png":"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\n"},"metadata":{}}],"source":["#4. Normalize the spectrograms and rotate them 90deg clockwise\n","# The mobile app running these models generates rotates spectrograms faster, and that's where every ms counts.\n","def normalize_and_permute(spectrograms, epsilon=1e-12):\n","    # mean = torch.mean(spectrograms, dim=(1, 2), keepdim=True)\n","    # std = torch.std(spectrograms, dim=(1, 2), keepdim=True) # tensorflow does not use Bessel's correction, Pytorch uses it by default\n","    # normalized = (spectrograms - mean) / (std + epsilon)\n","    # permuted = normalized.permute(0, 2, 1)\n","    permuted = spectrograms.permute(0, 2, 1)\n","    return torch.unsqueeze(permuted, dim=1)\n","\n","#Normalize and permute the training set, then plot the first spectrogram in the set\n","spectrogram_list = [item[\"spectrogram\"] for item in tagged_train_spectrograms]\n","train_spectrograms = normalize_and_permute(torch.tensor(spectrogram_list, dtype=torch.float32))\n","\n","print(train_spectrograms[0, 0, :, :].shape)\n","print(tagged_train_spectrograms[0][\"tags\"])\n","plt.figure(figsize=(12,12))\n","plt.imshow(train_spectrograms[0, 0, :, :], plt.get_cmap(\"gray\"))\n","plt.show()\n","\n","#Normalize and permute the test set, then plot the first spectrogram in the set\n","test_spectrogram_list = [item[\"spectrogram\"] for item in tagged_test_spectrograms]\n","test_spectrograms = normalize_and_permute(torch.tensor(test_spectrogram_list, dtype=torch.float32))\n","\n","print(test_spectrograms[0, 0, :, :].shape)\n","print(tagged_test_spectrograms[0][\"tags\"])\n","plt.figure(figsize=(12,12))\n","plt.imshow(test_spectrograms[0, 0, :, :], cmap=plt.get_cmap(\"gray\"))\n","plt.show()"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":1564,"status":"ok","timestamp":1680488324174,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"RF0REYPyiK_3","outputId":"23207e54-78e5-4ad1-b392-3aecaafa25d2"},"outputs":[{"output_type":"stream","name":"stdout","text":["['UNK', 'da2', 'gai4', 'ge1', 'hui2', 'qing3']\n","('END', 0), ('BEG', 1), ('BKG', 2), ('NOISE', 3), ('UNK', 4), ('da2', 5), ('gai4', 6), ('ge1', 7), ('hui2', 8), ('qing3', 9), "]}],"source":["#5. Extract all syllable names from the tracks\n","track_dir = colab_data + \"ac_audio/phrases/\"\n","def get_syllable_list():\n","    unique_syllables = set()\n","    for entry in os.listdir(track_dir):\n","        if entry.endswith(\".wav\"):\n","            track_name = entry.split(\".\")[0]\n","            syllables = track_name.split(\"_\")\n","            for s in syllables:\n","                unique_syllables.add(s)\n","\n","    return sorted(unique_syllables)\n","\n","syllables = get_syllable_list()\n","print(syllables)\n","\n","# Map all syllables to integer tokens\n","def map_syllables_to_tokens(syllables):\n","    syl_to_tokens = collections.OrderedDict()\n","    syl_to_tokens[\"END\"] = 0\n","    syl_to_tokens[\"BEG\"] = 1\n","    syl_to_tokens[\"BKG\"] = 2\n","    syl_to_tokens[\"NOISE\"] = 3\n","    for syllable in syllables:\n","        if syllable not in syl_to_tokens:\n","            syl_to_tokens[syllable] = len(syl_to_tokens)\n","\n","    return syl_to_tokens\n","\n","syl_to_tokens = map_syllables_to_tokens(syllables)\n","for item in syl_to_tokens.items():\n","    print(item, end=\", \")"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":816,"status":"ok","timestamp":1680488330318,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"D9hczgGzHlUK","outputId":"711450e8-32cd-47c9-ec25-bf8d9e253863"},"outputs":[{"output_type":"stream","name":"stdout","text":["2636\n","tensor([[1, 3, 0, 0, 0, 0],\n","        [1, 3, 0, 0, 0, 0],\n","        [1, 3, 0, 0, 0, 0]]) \n","\n","tensor([[3, 0, 0, 0, 0, 0],\n","        [3, 0, 0, 0, 0, 0],\n","        [3, 0, 0, 0, 0, 0]]) \n","\n","294\n","tensor([[1, 3, 0, 0, 0, 0],\n","        [1, 3, 0, 0, 0, 0],\n","        [1, 4, 7, 4, 0, 0]]) \n","\n","tensor([[3, 0, 0, 0, 0, 0],\n","        [3, 0, 0, 0, 0, 0],\n","        [4, 7, 4, 0, 0, 0]]) \n","\n"]}],"source":["#6. Extract the tag sequences from the tagged spectrograms, and convert them to sequences of integers\n","def get_tag_sequences(tagged_spectrograms, pad_to_length):\n","    tags = [item[\"tags\"].split(\"_\") for item in tagged_spectrograms]\n","    input_tokens = [[syl_to_tokens[\"BEG\"]] + [syl_to_tokens[tag] for tag in tag_array] for tag_array in tags]\n","    output_tokens = [[syl_to_tokens[tag] for tag in tag_array] + [syl_to_tokens[\"END\"]] for tag_array in tags]\n","\n","    padded_input_tokens = torch.zeros((len(tags), pad_to_length), dtype=torch.int64)\n","    padded_output_tokens = torch.zeros((len(tags), pad_to_length), dtype=torch.int64)\n","\n","    for idx, tag_array in enumerate(input_tokens):\n","        padded_input_tokens[idx, :len(tag_array)] = torch.tensor(tag_array, dtype=torch.int64)\n","\n","    for idx, tag_array in enumerate(output_tokens):\n","        padded_output_tokens[idx, :len(tag_array)] = torch.tensor(tag_array, dtype=torch.int64)\n","\n","    return padded_input_tokens, padded_output_tokens\n","\n","#Convert both training and test set tags into input + output tags, printing a few samples of each set\n","max_train_tags = np.amax([len(item[\"tags\"].split(\"_\")) for item in tagged_train_spectrograms])\n","max_test_tags = np.amax([len(item[\"tags\"].split(\"_\")) for item in tagged_test_spectrograms])\n","pad_to_length = max(max_train_tags, max_test_tags) + 1\n","\n","train_input_tokens, train_output_tokens = get_tag_sequences(tagged_train_spectrograms, pad_to_length)\n","test_input_tokens, test_output_tokens = get_tag_sequences(tagged_test_spectrograms, pad_to_length)\n","\n","print(len(train_input_tokens))\n","print(train_input_tokens[:3], \"\\n\")\n","print(train_output_tokens[:3], \"\\n\")\n","\n","print(len(test_input_tokens))\n","print(test_input_tokens[:3], \"\\n\")\n","print(test_output_tokens[:3], \"\\n\")"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"yWnOshJsrhqS"},"outputs":[],"source":["#7. Place all tensors on the GPU, if available\n","train_spectrograms = train_spectrograms.to(device)\n","test_spectrograms = test_spectrograms.to(device)\n","train_input_tokens = train_input_tokens.to(device)\n","train_output_tokens = train_output_tokens.to(device)\n","test_input_tokens = test_input_tokens.to(device)\n","test_output_tokens = test_output_tokens.to(device)\n","\n","# Network hyper params\n","freq_features = train_spectrograms[0, 0, :, :].shape[1]\n","conv_channels = 10\n","conv_kernel_size = (5, 1)\n","conv_padding_size = (conv_kernel_size[0] // 2, conv_kernel_size[1] // 2)\n","rnn_hidden_size = 64\n","rnn_stacked_layers = 2\n","embedding_dim = math.ceil(2/3*len(syl_to_tokens))"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":2977,"status":"ok","timestamp":1680488348672,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"GRsn52q1KQLG","outputId":"52f2f065-e255-4e8a-9ac2-3483e27b5ea3"},"outputs":[{"output_type":"stream","name":"stdout","text":["VoxEncoder(\n","  (conv_1): Conv2d(1, 10, kernel_size=(5, 1), stride=(1, 1), padding=(2, 0))\n","  (relu_1): LeakyReLU(negative_slope=0.01)\n","  (bn_1): BatchNorm2d(10, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","  (conv_2): Conv2d(10, 1, kernel_size=(5, 1), stride=(1, 1), padding=(2, 0))\n","  (relu_2): LeakyReLU(negative_slope=0.01)\n","  (bn_2): BatchNorm2d(1, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","  (encoder): GRU(80, 64, num_layers=2, batch_first=True, bidirectional=True)\n",")\n","VoxDecoder(\n","  (tag_embed): Embedding(10, 7)\n","  (decoder): GRU(7, 128, num_layers=2, batch_first=True)\n","  (align): Softmax(dim=-1)\n","  (linear_1): Linear(in_features=256, out_features=64, bias=True)\n","  (tanh_1): Tanh()\n","  (linear_2): Linear(in_features=64, out_features=10, bias=True)\n","  (tanh_2): Tanh()\n",")\n","130693\n","168848\n","299541\n"]}],"source":["#8. The model is an sequence-to-sequence model, made of an encoder and a decoder.\n","# This is the encoder...\n","class VoxEncoder(torch.nn.Module):\n","    def __init__(self):\n","        #Init constants\n","        super(VoxEncoder, self).__init__()\n","\n","        #Layers\n","        self.conv_1 = nn.Conv2d(in_channels=1, out_channels=conv_channels, kernel_size=conv_kernel_size, padding=conv_padding_size)\n","        self.relu_1 = nn.LeakyReLU()\n","        self.bn_1 = nn.BatchNorm2d(num_features=conv_channels) # num_features = out_channels in the conv layer\n","\n","        self.conv_2 = nn.Conv2d(in_channels=conv_channels, out_channels=1, kernel_size=conv_kernel_size, padding=conv_padding_size)\n","        self.relu_2 = nn.LeakyReLU()\n","        self.bn_2 = nn.BatchNorm2d(num_features=1)\n","\n","        self.encoder = nn.GRU(input_size=freq_features, hidden_size=rnn_hidden_size, batch_first=True, bidirectional=True, num_layers=rnn_stacked_layers)\n","\n","    def forward(self, spectrogram):\n","        #Conv block 1\n","        x = spectrogram\n","        x = self.conv_1(x)\n","        x = self.relu_1(x)\n","        x = self.bn_1(x)\n","\n","        #Conv block 2\n","        x = self.conv_2(x)\n","        x = self.relu_2(x)\n","        x = self.bn_2(x)\n","\n","        #Encoder\n","        x = torch.squeeze(x, dim=1)\n","        encoder_output, enc_h = self.encoder(x)\n","\n","        #Reshape hidden state to match the dimensions expected by the decoder\n","        h_layers_twice, h_batch_size, h_output_dim = enc_h.shape\n","        unidirectional_h = enc_h.reshape((h_layers_twice // 2, h_batch_size, h_output_dim * 2))\n","\n","        return encoder_output, unidirectional_h\n","\n","#... And this is the decoder.\n","class VoxDecoder(torch.nn.Module):\n","    def __init__(self):\n","        #Init constants\n","        super(VoxDecoder, self).__init__()\n","\n","        #Layers\n","        self.tag_embed = nn.Embedding(num_embeddings=len(syl_to_tokens), embedding_dim=embedding_dim)\n","        self.decoder = nn.GRU(input_size=embedding_dim, hidden_size=2*rnn_hidden_size, batch_first=True, bidirectional=False, num_layers=rnn_stacked_layers)\n","\n","        self.align = nn.Softmax(dim=-1)\n","\n","        self.linear_1 = nn.Linear(4*rnn_hidden_size, rnn_hidden_size)\n","        self.tanh_1 = nn.Tanh()\n","\n","        self.linear_2 = nn.Linear(rnn_hidden_size, len(syl_to_tokens))\n","        self.tanh_2 = nn.Tanh()\n","\n","    #When using teacher forcing (training phase), tags will be of shape (batch, tag_len)\n","    #When not using teacher forcing (test phase), tags will be of shape (batch, 1)  (essentially [[\"BEG\"], [\"BEG\"], ...]\n","    def forward(self, encoder_output, hidden_state, tags):\n","        #Decoder\n","        tag_embeddings = self.tag_embed(tags)\n","        decoder_output, dec_h = self.decoder(tag_embeddings, hidden_state)\n","\n","        #Attention block\n","        attention_scores = torch.bmm(decoder_output, encoder_output.permute(0, 2, 1)) #bmm = batch matrix multiplication\n","        alignment_vectors = self.align(attention_scores)\n","        context_vectors = torch.bmm(alignment_vectors, encoder_output)\n","        concat_contexts = torch.cat([context_vectors, decoder_output], dim=-1)\n","\n","        #Time-distributed linear (dense) layers\n","        x = self.linear_1(concat_contexts)\n","        x = self.tanh_1(x)\n","        x = self.linear_2(x)\n","        output = self.tanh_2(x)\n","\n","        return output, dec_h, alignment_vectors\n","\n","#Instantiate the model and place on the GPU, if available\n","encoder = VoxEncoder().to(device)\n","print(encoder)\n","\n","decoder = VoxDecoder().to(device)\n","print(decoder)\n","\n","# Print the parameter count for the encoder and the decoder\n","encoder_param_count = sum(param.numel() for param in encoder.parameters() if param.requires_grad)\n","decoder_param_count = sum(param.numel() for param in decoder.parameters() if param.requires_grad)\n","\n","print(encoder_param_count)\n","print(decoder_param_count)\n","print(encoder_param_count + decoder_param_count)"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"oG84ReFL_UWN"},"outputs":[],"source":["# Keeping this cell from the tensorflow version... Can't you plot pytorch models?\n","# This is how the full model actually looks like."]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":3827,"status":"ok","timestamp":1680488360826,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"hsLqRV9y0EWC","outputId":"af987155-e82b-4282-fb1d-6bd58eb174ec"},"outputs":[{"output_type":"stream","name":"stdout","text":["0.08460995265471817\n","0.9153900473452818\n"]},{"output_type":"stream","name":"stderr","text":["<ipython-input-9-590b6796d869>:13: DeprecationWarning: `np.float` is a deprecated alias for the builtin `float`. To silence this warning, use `float` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.float64` here.\n","Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n","  raw_token_weights = np.array(list(token_frequency.values()), dtype=np.float)\n"]}],"source":["#9. Calculate some helper params for the weighted loss and accuracy functions\n","\n","# Run sample data through the model, to visualize the output and it's dimensions\n","encoder_output, encoder_state = encoder(test_spectrograms[:1])\n","decoder_output, decoder_state, align_vec = decoder(encoder_output, encoder_state, test_input_tokens[:1])\n","#print(decoder_output, decoder_output.shape)\n","\n","token_frequency = {token:0 for token in syl_to_tokens.values()}\n","for output_line in train_output_tokens:\n","    for token in output_line:\n","        token_frequency[token.item()] += 1\n","\n","raw_token_weights = np.array(list(token_frequency.values()), dtype=np.float)\n","inverse_token_weights = np.where(raw_token_weights == 0, 1.0, 1 / (raw_token_weights + 1e-12))\n","token_weights = torch.tensor(inverse_token_weights / np.sum(inverse_token_weights), dtype=torch.float32).to(device)\n","loss_scale = torch.mean(torch.tensor(raw_token_weights, dtype=torch.float32)).item()\n","\n","onehot_tokens = torch.eye(len(token_weights), dtype=torch.float32).to(device)\n","\n","# Multi-label categorical crossentropy loss on one-hot output vectors, weighted by the frequency of each token in the training set\n","def weighted_sparse_categorical_crossentropy(labels, logits):\n","    softmax = F.softmax(logits, dim=-1)\n","    onehot_labels = torch.stack([torch.stack([onehot_tokens[token.item()] for token in sequence]) for sequence in labels])\n","    scalar_cce = - onehot_labels * torch.log(softmax + 1e-12)\n","\n","    reduced_cce = torch.mean(torch.sum(scalar_cce * token_weights, dim=-1, keepdim=True))\n","    return loss_scale * reduced_cce\n","\n","weighted_sparse_categorical_crossentropy(test_output_tokens[:2, :1], decoder_output)\n","\n","# Multi-label categorical accuracy metric on one-hot output vectors, weighted by the frequency of each token in the training set\n","class WeightedSparseCategoricalAccuracy():\n","    def __init__(self, token_weights, from_logits=True):\n","        self.accurate_sum = 0.0\n","        self.total_sum = 0.0\n","        self.token_weights = token_weights\n","        self.from_logits = from_logits\n","\n","    def update_state(self, y_true, y_pred, sample_weight=None):\n","        predicted_tokens = torch.argmax(y_pred, dim=-1, keepdim=True).squeeze(dim=-1) if self.from_logits else y_pred\n","        gather_weights = self.token_weights[y_true] #indexing trick from https://stackoverflow.com/questions/53697596/pytorch-equivalent-tf-gather\n","        accurate_weights = torch.sum(torch.where(y_true == predicted_tokens, gather_weights, torch.tensor(0.0, dtype=gather_weights.dtype).to(device)))\n","        total_weights = torch.sum(gather_weights)\n","\n","        self.accurate_sum += accurate_weights.item()\n","        self.total_sum += total_weights.item()\n","\n","    def result(self):\n","        return self.accurate_sum / self.total_sum\n","\n","    def reset_states(self):\n","        self.accurate_sum = 0.0\n","        self.total_sum = 0.0\n","\n","# Calculate our weighted metric on two example pairs of tokens\n","wsca = WeightedSparseCategoricalAccuracy(token_weights=token_weights, from_logits=False)\n","\n","wsca.update_state(\n","    torch.tensor([[5, 0, 0, 0]], dtype=torch.long).to(device), \n","    torch.tensor([[3, 0, 0, 0]], dtype=torch.long).to(device)\n",")\n","print(wsca.result())\n","wsca.reset_states()\n","\n","wsca.update_state(\n","    torch.tensor([[5, 0, 0, 0]], dtype=torch.long).to(device), \n","    torch.tensor([[5, 1, 1, 1]], dtype=torch.long).to(device)\n",")\n","print(wsca.result())\n","wsca.reset_states()"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"upC_Jo4sPC2-"},"outputs":[],"source":["#10. Training step for a batch of spectrograms + tags, using teacher forcing\n","def train_batch_tf(encoder, decoder, spectrograms, input_tokens, output_tokens, encoder_optimizer, decoder_optimizer):\n","    encoder.train()\n","    decoder.train()\n","    encoder_optimizer.zero_grad()\n","    decoder_optimizer.zero_grad()\n","\n","    encoder_output, encoder_state = encoder(spectrograms)\n","    hidden_state = encoder_state\n","\n","    decoder_iterations = output_tokens.shape[-1]\n","    predictions = []\n","    for iter in range(decoder_iterations):\n","        output_logits, hidden_state, _ = decoder(encoder_output, hidden_state, input_tokens[:, iter:iter+1])\n","        predictions.append(output_logits)\n","\n","    predictions = torch.cat(predictions, dim=1)\n","    cross_entropy_loss = weighted_sparse_categorical_crossentropy(output_tokens, predictions)\n","    cross_entropy_loss.backward()\n","    \n","    decoder_optimizer.step()\n","    encoder_optimizer.step()\n","\n","    return cross_entropy_loss.item(), predictions\n","\n","# Training step for a batch of spectrograms + tags, without teacher forcing\n","def train_batch(encoder, decoder, spectrograms, output_tokens, encoder_optimizer, decoder_optimizer):\n","    encoder.train()\n","    decoder.train()\n","    encoder_optimizer.zero_grad()\n","    decoder_optimizer.zero_grad()\n","\n","    encoder_output, encoder_state = encoder(spectrograms)\n","    hidden_state = encoder_state\n","    input_tokens = torch.full(\n","        (output_tokens.shape[0], 1), \n","        syl_to_tokens[\"BEG\"]\n","    ).to(device)\n","\n","    decoder_iterations = output_tokens.shape[-1]\n","    predictions = []\n","    for iter in range(decoder_iterations):\n","        output_logits, hidden_state, _ = decoder(encoder_output, hidden_state, input_tokens)\n","        _, predicted_tokens = output_logits.topk(1)\n","        input_tokens = predicted_tokens.squeeze(-1).detach()\n","        predictions.append(output_logits)\n","\n","    predictions = torch.cat(predictions, dim=1)\n","    cross_entropy_loss = weighted_sparse_categorical_crossentropy(output_tokens, predictions)\n","    cross_entropy_loss.backward()\n","    \n","    decoder_optimizer.step()\n","    encoder_optimizer.step()\n","\n","    return cross_entropy_loss.item(), predictions\n","\n","# Evaluation step for a batch of spectrograms + tags, using teacher forcing\n","def test_batch_tf(encoder, decoder, spectrograms, input_tokens, output_tokens):\n","    encoder.eval()\n","    decoder.eval()\n","\n","    encoder_output, encoder_state = encoder(spectrograms)\n","    hidden_state = encoder_state\n","\n","    decoder_iterations = output_tokens.shape[-1]\n","    predictions = []\n","    for iter in range(decoder_iterations):\n","        output_logits, hidden_state, _ = decoder(encoder_output, hidden_state, input_tokens[:, iter:iter+1])\n","        predictions.append(output_logits)\n","\n","    predictions = torch.cat(predictions, dim=1)\n","    cross_entropy_loss = weighted_sparse_categorical_crossentropy(output_tokens, predictions)\n","\n","    return cross_entropy_loss.item(), predictions\n","\n","# Evaluation step for a batch of spectrograms + tags, without teacher forcing\n","def test_batch(encoder, decoder, spectrograms, output_tokens, log=False, force_cpu=False):\n","    encoder.eval()\n","    decoder.eval()\n","\n","    encoder_output, encoder_state = encoder(spectrograms)\n","    hidden_state = encoder_state\n","    input_tokens = torch.full(\n","        (output_tokens.shape[0], 1), \n","        syl_to_tokens[\"BEG\"]\n","    ).to(cpu if force_cpu else device)\n","\n","    if log:\n","        print(\"encoder_output: \", encoder_output, encoder_output.shape)\n","        print(\"encoder_state: \", encoder_state, encoder_state.shape)\n","        print(\"\\n\")\n","\n","    decoder_iterations = output_tokens.shape[-1]\n","    predictions = []\n","    for iter in range(decoder_iterations):\n","        output_logits, hidden_state, _ = decoder(encoder_output, hidden_state, input_tokens)\n","        if log:\n","            print(\"output_logits: \", output_logits, output_logits.shape)\n","            print(\"hidden_state: \", hidden_state, hidden_state.shape)\n","            print(\"\\n\")\n","\n","        _, predicted_tokens = output_logits.topk(1)\n","        input_tokens = predicted_tokens.squeeze(-1).detach()\n","        predictions.append(output_logits)\n","\n","    predictions = torch.cat(predictions, dim=1)\n","    cross_entropy_loss = weighted_sparse_categorical_crossentropy(output_tokens, predictions)\n","\n","    return cross_entropy_loss.item(), predictions\n","\n","# Train the model for the specified number of epochs\n","def training_loop(encoder, decoder, batch_size, epochs=1, teacher_forcing_ratio=0.5):\n","    train_examples = len(train_spectrograms)\n","    test_examples = len(test_spectrograms)\n","    train_batches = math.ceil(train_examples / batch_size)\n","    test_batches = math.ceil(test_examples / batch_size)\n","\n","    encoder_optimizer = torch.optim.Adam(encoder.parameters(), lr=0.001)\n","    decoder_optimizer = torch.optim.Adam(decoder.parameters(), lr=0.001)\n","    train_accuracy = WeightedSparseCategoricalAccuracy(token_weights=token_weights)\n","    test_accuracy = WeightedSparseCategoricalAccuracy(token_weights=token_weights)\n","\n","    print(\"Starting training for {} epochs...\".format(epochs))\n","    for epoch in range(epochs):\n","        epoch_start = time.time()\n","        for batch_idx in range(train_batches):\n","            batch_start = batch_idx * batch_size\n","            batch_end = min(batch_start + batch_size, train_examples)\n","\n","            spectrogram_batch = train_spectrograms[batch_start:batch_end]\n","            input_tokens_batch = train_input_tokens[batch_start:batch_end]\n","            output_tokens_batch = train_output_tokens[batch_start:batch_end]\n","\n","            if np.random.rand() >= teacher_forcing_ratio:\n","                batch_loss, logits = train_batch_tf(encoder, decoder, spectrogram_batch, input_tokens_batch, output_tokens_batch, encoder_optimizer, decoder_optimizer)\n","            else:\n","                batch_loss, logits = train_batch(encoder, decoder, spectrogram_batch, output_tokens_batch, encoder_optimizer, decoder_optimizer)\n","            train_accuracy.update_state(output_tokens_batch, logits)\n","\n","        for batch_idx in range(test_batches):\n","            batch_start = batch_idx * batch_size\n","            batch_end = min(batch_start + batch_size, train_examples)\n","\n","            spectrogram_batch = test_spectrograms[batch_start:batch_end]\n","            #input_tokens_batch = test_input_tokens[batch_start:batch_end]\n","            output_tokens_batch = test_output_tokens[batch_start:batch_end]\n","\n","            batch_loss, logits = test_batch(encoder, decoder, spectrogram_batch, output_tokens_batch)\n","            test_accuracy.update_state(output_tokens_batch, logits)\n","\n","        print(\"\\nModel metrics for epoch {}: \".format(epoch + 1))\n","        print(\"Test Loss: {}\".format(batch_loss))\n","        print(\"Weighted Train Accuracy: {}\".format(train_accuracy.result()))\n","        print(\"Weighted Test Accuracy: {}\".format(test_accuracy.result()))\n","        print (\"Training time was {} sec\".format(time.time() - epoch_start))\n","\n","        train_accuracy.reset_states()\n","        test_accuracy.reset_states()\n","\n","#Experiment with hyperparams\n","# conv_channels = 10\n","# conv_kernel_size = (5, 1)\n","# rnn_hidden_size = 64\n","# rnn_stacked_layers = 2\n","\n","# encoder = VoxEncoder().to(device)\n","# decoder = VoxDecoder().to(device)\n","training_loop(\n","    encoder,\n","    decoder,\n","    batch_size=1024, \n","    epochs=800\n",")\n","\n","# Save the full model, for future use or inference\n","torch.save(encoder.state_dict(), colab_data + \"full_vox_encoder.pt\")\n","torch.save(decoder.state_dict(), colab_data + \"full_vox_decoder.pt\")\n","\n","# Save the model as a PyTorch Native C++ module, for use in the Android App\n","encoder = encoder.to(cpu).eval()\n","decoder = decoder.to(cpu).eval()\n","\n","mock_encoder_input = torch.tensor(np.random.rand(*test_spectrograms[:1].shape), dtype=torch.float)\n","mock_encoder_output = torch.tensor(np.random.rand(*encoder_output.shape), dtype=torch.float)\n","mock_encoder_state = torch.tensor(np.random.rand(*encoder_state.shape), dtype=torch.float)\n","mock_decoder_input = torch.tensor(np.random.rand(*test_input_tokens[:1].shape), dtype=torch.long)\n","\n","traced_encoder = torch.jit.trace(encoder, mock_encoder_input)\n","traced_decoder = torch.jit.trace(decoder, (mock_encoder_output, mock_encoder_state, mock_decoder_input))\n","\n","traced_encoder.save(colab_data + \"vox_encoder.pt\")\n","traced_decoder.save(colab_data + \"vox_decoder.pt\")\n","\n","encoder = encoder.to(device)\n","decoder = decoder.to(device)"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"dOm1KF79zrLn"},"outputs":[],"source":["# Load the trained model from storage, for quick evaluations\n","# encoder.load_state_dict(torch.load(colab_data + \"full_vox_encoder.pt\"))\n","# decoder.load_state_dict(torch.load(colab_data + \"full_vox_decoder.pt\"))\n","\n","# encoder = encoder.to(device).eval()\n","# decoder = decoder.to(device).eval()"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":52,"status":"ok","timestamp":1673625180549,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"2PG0mO7-AW6z","outputId":"81577517-586b-4f1f-ff3b-a835165ddf9c"},"outputs":[{"name":"stdout","output_type":"stream","text":["encoder_output:  tensor([[[ 0.2032,  0.0018, -0.0682,  ...,  0.1416,  0.6037,  0.4808],\n","         [ 0.2585, -0.8097, -0.7170,  ...,  0.9819,  0.9878, -0.2247],\n","         [ 0.2472, -0.6320, -0.7592,  ...,  0.9813,  0.9975, -0.7735],\n","         ...,\n","         [ 0.8572,  0.9431,  0.9709,  ...,  0.3704,  0.9617, -0.5688],\n","         [ 0.8148,  0.9334,  0.9703,  ...,  0.3093,  0.9121, -0.2953],\n","         [ 0.8110,  0.9239,  0.9698,  ...,  0.2065,  0.7724,  0.1343]]],\n","       device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([1, 138, 128])\n","encoder_state:  tensor([[[-0.9716, -0.2954, -0.3430, -0.8391,  0.9995,  0.7539, -0.4577,\n","          -0.1445,  0.9653, -0.9565,  0.9788, -0.9714,  0.8916, -0.8213,\n","          -0.5300, -0.7090,  0.3092,  0.9441,  0.9765, -0.9386, -0.9861,\n","          -0.9995,  0.9589,  0.3641,  0.9201, -0.9849,  0.9305, -0.9287,\n","           0.9240,  0.9966,  0.9995,  0.9483, -0.9473,  0.9488,  0.4819,\n","          -0.9737, -0.8393, -0.9954, -0.9696,  0.8701,  0.7906,  0.9990,\n","           0.9286, -0.9680,  0.3284, -0.8321, -0.9902, -0.9987, -0.5447,\n","          -0.8352,  0.9983, -0.9995,  0.8316,  0.6319, -0.9518, -0.9941,\n","          -0.9269,  0.7521,  0.7651, -0.9896,  0.9981, -0.6281,  0.9052,\n","          -0.1211, -0.9638, -0.8216,  0.7355, -0.1772,  0.8788, -0.3760,\n","           0.2359,  0.8041,  0.6882, -0.2462, -0.9736,  0.3004,  0.9767,\n","           0.2107,  0.7367,  0.9809,  0.8386,  0.6404, -0.6318, -0.6317,\n","          -0.7138,  0.7823,  0.9815, -0.3013,  0.9437, -0.9808, -0.9763,\n","          -0.8875,  0.6629,  0.1099, -0.9326,  0.8015,  0.6177, -0.7478,\n","           0.7556,  0.6864, -0.8956,  0.8988,  0.2321,  0.3686, -0.1107,\n","          -0.8966, -0.9648, -0.2411,  0.8834,  0.7048, -0.3900, -0.4699,\n","          -0.2678, -0.0585, -0.8756, -0.0022, -0.7394,  0.5366, -0.6195,\n","           0.9530,  0.8131, -0.6102,  0.9587, -0.9154,  0.2939,  0.7076,\n","           0.1056, -0.6928]],\n","\n","        [[ 0.8110,  0.9239,  0.9698,  0.5465,  0.9704,  0.2142,  0.6989,\n","          -0.1449,  0.9780,  0.9307, -0.2232,  0.3803,  0.9567, -0.9418,\n","          -0.9851, -0.9953, -0.8651, -0.9518,  0.9788, -0.9442, -0.6924,\n","           0.9998, -0.4531, -0.9970, -0.8811,  0.9994,  0.9887,  0.1072,\n","           0.9993, -0.8995, -0.9891,  0.8257,  0.5541,  0.4217, -0.9836,\n","          -0.2638, -0.4649, -0.9084, -0.9210,  0.9910,  0.4559,  0.9999,\n","          -0.7667, -0.6024, -0.8094,  0.4741, -0.9417,  0.9985,  0.0322,\n","           0.9092,  0.8939,  0.6675,  0.9984,  0.7339,  0.9813,  0.9983,\n","           0.2473,  0.9991,  0.9815,  0.9927, -0.8927, -0.8641, -0.4718,\n","          -0.8548, -0.9378, -0.1089, -0.3131,  0.9620, -0.4164, -0.6372,\n","           0.5487,  0.8828,  0.4291, -0.8114,  0.8421, -0.8944, -0.9794,\n","           0.6517,  0.5711, -0.9320,  0.6249, -0.9506,  0.8605,  0.8821,\n","           0.2839,  0.5484,  0.6965, -0.6904, -0.3193, -0.9016,  0.0591,\n","          -0.8987, -0.8796, -0.3800,  0.9898, -0.9452, -0.9613,  0.4172,\n","           0.9192,  0.2768, -0.8158, -0.9864, -0.8895,  0.9882,  0.5325,\n","          -0.1393, -0.1973, -0.9448, -0.4554,  0.8231, -0.3226,  0.7860,\n","           0.8247,  0.6207, -0.5572,  0.8635,  0.9756, -0.5100,  0.9603,\n","          -0.1848, -0.9685, -0.1204,  0.8685, -0.8216,  0.1176,  0.1416,\n","           0.6037,  0.4808]]], device='cuda:0',\n","       grad_fn=<ReshapeAliasBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.9998, -1.0000, -1.0000, -0.9925, -1.0000,  0.9981, -0.9947,\n","          -1.0000, -1.0000]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-7.8823e-01, -1.4283e-01, -3.3608e-01, -5.4403e-01,  8.3079e-01,\n","          -3.3556e-01,  1.4513e-01, -3.6031e-02,  7.7715e-01, -5.5364e-01,\n","           4.3400e-01, -3.0284e-01,  8.9519e-01, -6.0824e-01, -1.3675e-01,\n","          -6.9446e-01,  6.6219e-05,  6.1138e-01,  6.3767e-02,  3.7167e-01,\n","          -7.2228e-01, -7.9999e-01,  8.6967e-01,  6.7817e-01,  5.5981e-01,\n","          -5.0007e-01,  5.5678e-01, -8.2868e-01,  6.2290e-01,  8.2575e-01,\n","           6.0486e-01,  3.9435e-01, -1.6985e-01,  3.4769e-01, -1.5269e-01,\n","          -9.0400e-01, -6.4265e-01, -7.9287e-01, -6.1337e-01,  5.9224e-01,\n","           1.8822e-01,  8.3710e-01, -3.8968e-01,  1.0150e-01,  2.4829e-01,\n","           1.7091e-01, -8.5459e-01, -5.9687e-01,  8.7980e-02,  3.6573e-01,\n","           8.7928e-01, -9.3552e-01, -2.6949e-01,  5.4400e-01,  1.2005e-01,\n","           1.7616e-01, -1.8199e-01, -1.0497e-01,  5.8708e-01,  7.1217e-02,\n","          -1.1719e-01, -8.1170e-01,  4.6673e-01,  3.9112e-01, -5.6092e-01,\n","           4.3219e-01,  3.3144e-01, -4.0515e-01,  6.3270e-01,  1.1666e-01,\n","           4.3907e-01,  5.4415e-01, -1.8400e-01, -5.0116e-01, -9.1032e-01,\n","           4.9010e-01,  5.3854e-01,  4.7230e-01,  2.8820e-01,  2.9715e-01,\n","           4.9797e-01,  6.2261e-01, -2.0871e-01, -2.9418e-01, -4.5979e-01,\n","           6.9615e-01,  4.7777e-02, -3.9341e-02,  7.0188e-01, -3.3003e-01,\n","          -2.2650e-01,  2.6054e-01,  8.5935e-01,  6.6162e-01, -2.2991e-01,\n","          -7.8452e-02,  6.1598e-01, -4.8259e-01, -3.5695e-01,  3.2232e-01,\n","          -8.1255e-01,  4.4961e-01, -1.0458e-02, -5.4665e-02, -1.7327e-01,\n","          -3.0429e-01, -6.0588e-01,  1.2828e-03,  2.6767e-01,  2.0649e-01,\n","          -6.6393e-01, -2.5846e-01,  2.0615e-01,  2.1996e-01, -7.6027e-03,\n","          -4.9003e-01, -6.2890e-01,  2.6964e-01,  1.3340e-01,  4.2630e-01,\n","           4.7848e-01, -4.4969e-01,  8.2957e-01, -9.0880e-01,  7.9633e-02,\n","           1.3310e-01, -1.4812e-01, -5.3197e-01]],\n","\n","        [[ 9.2858e-01, -8.3968e-01, -6.7377e-01,  1.3604e-01, -1.8080e-01,\n","           9.6772e-01,  4.8742e-01, -2.7545e-03,  9.4010e-01,  6.3026e-01,\n","          -3.5133e-01, -2.7430e-02, -6.8170e-01, -6.5562e-01,  9.4529e-01,\n","           7.2815e-01,  1.7030e-01, -9.5493e-01, -9.3706e-01,  9.6651e-01,\n","           3.9444e-01, -9.5736e-01, -4.3107e-01,  9.6090e-01,  9.7381e-01,\n","          -9.4700e-01, -8.1372e-01, -8.9711e-01, -8.3862e-01,  7.8972e-01,\n","           6.6579e-01,  6.0933e-01,  6.2030e-01,  5.8658e-01,  9.1848e-01,\n","          -5.3833e-01, -4.7687e-01, -9.2769e-01,  4.3485e-01, -8.3130e-01,\n","          -8.2720e-01, -9.4322e-01, -6.6048e-01, -5.9881e-01,  7.7983e-01,\n","          -2.2370e-01, -7.5899e-01, -9.6914e-01,  5.4931e-01,  7.2452e-01,\n","          -9.5024e-01, -9.3129e-01, -8.3284e-01, -2.0672e-01,  7.7051e-01,\n","          -9.5722e-01,  8.0453e-01, -9.6032e-01, -9.6028e-01, -9.6082e-01,\n","           4.2388e-01, -5.9720e-01, -6.3031e-01, -6.6991e-01, -3.4503e-02,\n","           7.4126e-01, -9.3790e-01,  9.7220e-01, -4.4410e-01,  4.1539e-01,\n","           4.5994e-01, -5.8898e-01,  8.2649e-01,  6.7877e-01, -6.2258e-01,\n","          -8.6005e-01, -8.2817e-01, -6.5626e-01,  7.6110e-01, -4.3821e-01,\n","           4.2680e-01, -9.4565e-01,  5.4344e-01, -6.6735e-01, -2.3261e-01,\n","           6.6043e-01,  5.8538e-01, -6.6767e-01, -2.7084e-01,  8.5898e-01,\n","           4.6020e-01, -4.6031e-01, -8.0112e-01,  8.8322e-01,  9.7681e-01,\n","          -8.4475e-01,  7.9323e-01,  5.7988e-01, -8.7562e-01, -2.6758e-01,\n","          -8.2493e-01, -8.5860e-01, -3.6836e-01,  7.1213e-01,  2.8881e-01,\n","           8.7556e-01,  9.4906e-01, -8.9617e-01,  9.0405e-01,  3.2253e-01,\n","           7.9511e-01,  7.2577e-01, -6.9813e-01,  9.5809e-01, -4.3745e-01,\n","           9.8111e-01,  5.1705e-01, -5.2230e-01,  9.8956e-01, -1.8945e-01,\n","          -2.9881e-01, -1.6758e-01,  7.8235e-01, -9.5211e-01,  7.8760e-01,\n","           5.7624e-01, -7.1810e-01,  4.1148e-01]]], device='cuda:0',\n","       grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.9995, -1.0000, -1.0000, -1.0000, -1.0000,  1.0000,  0.9977,\n","          -0.9999, -1.0000]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.5933,  0.2560,  0.1293, -0.2648,  0.5301,  0.1327,  0.5109,\n","           0.0389,  0.6916, -0.5026,  0.6266, -0.1623,  0.4151, -0.4867,\n","          -0.2347, -0.1968,  0.3234, -0.0459, -0.0230,  0.1902, -0.1254,\n","           0.2976,  0.3653,  0.8259,  0.4758, -0.1819,  0.3605, -0.8472,\n","           0.5683,  0.8111, -0.0947,  0.0261, -0.0641,  0.0564, -0.0522,\n","           0.1317, -0.1756,  0.2776, -0.3393, -0.2807,  0.0197,  0.5861,\n","          -0.4474,  0.0898, -0.1680,  0.0805, -0.7930,  0.5162,  0.6372,\n","           0.1763,  0.5356,  0.0881, -0.2905,  0.2694, -0.0650,  0.0184,\n","          -0.1133,  0.1595,  0.5532,  0.2057,  0.0331, -0.2086,  0.0783,\n","           0.6479, -0.1871, -0.2523,  0.4434,  0.0020,  0.0163,  0.5177,\n","           0.1671,  0.3415, -0.4470, -0.5470, -0.4948,  0.2399, -0.2391,\n","           0.7367,  0.4680, -0.0950,  0.5719,  0.0650, -0.2765, -0.6680,\n","          -0.0787,  0.2986, -0.2950,  0.5379, -0.0699, -0.0834, -0.1580,\n","           0.3438,  0.4306,  0.1475,  0.1543, -0.4320,  0.1406, -0.5703,\n","          -0.3149, -0.6127, -0.6156, -0.3443, -0.2591, -0.2913, -0.1996,\n","          -0.3249, -0.3405,  0.0278, -0.3517, -0.1012, -0.5961,  0.0459,\n","          -0.1607, -0.6417,  0.1468, -0.2488, -0.2015, -0.4007, -0.2921,\n","           0.3901, -0.4468, -0.1233,  0.5986, -0.7618, -0.4201,  0.3751,\n","          -0.5106, -0.5309]],\n","\n","        [[-0.6476, -0.4694, -0.5421, -0.3238,  0.8607, -0.9733,  0.5154,\n","          -0.7235, -0.9959,  0.4755,  0.9318, -0.6829, -0.4910,  0.5660,\n","          -0.8501,  0.2868,  0.2678,  0.7176, -0.8157, -0.9592,  0.5254,\n","          -0.9762, -0.4371, -0.6228, -0.8011,  0.2697, -0.9241, -0.1988,\n","          -0.3259, -0.0674, -0.9737,  0.9360,  0.3955,  0.5309, -0.9515,\n","           0.8158,  0.8148,  0.9235,  0.2577, -0.9718, -0.9230, -0.7920,\n","          -0.4727,  0.0169,  0.7138, -0.3228, -0.8687,  0.2221,  0.6212,\n","          -0.5545, -0.1723, -0.8922, -0.2720, -0.9200,  0.9461,  0.9729,\n","           0.6417, -0.9654, -0.9203,  0.9610, -0.8959, -0.9713, -0.9069,\n","           0.4214,  0.2620,  0.3038, -0.8741,  0.9793, -0.6478,  0.8543,\n","          -0.0778,  0.9168, -0.1198,  0.4595,  0.9773,  0.4134, -0.9105,\n","           0.8936,  0.9364, -0.6810,  0.9374, -0.5406,  0.9246,  0.9540,\n","          -0.4675,  0.9270, -0.2291, -0.6221,  0.1929,  0.4137, -0.8497,\n","          -0.8463, -0.9827,  0.9245,  0.9577, -0.7752, -0.9777,  0.6209,\n","           0.8646,  0.8264, -0.8290, -0.6692, -0.8012,  0.9622,  0.3452,\n","          -0.9304,  0.5166,  0.4038, -0.9818,  0.9803,  0.8614,  0.3654,\n","           0.8256,  0.8582, -0.9810,  0.6761,  0.5671,  0.1422,  0.7540,\n","          -0.9844, -0.9465, -0.8949,  0.9816, -0.8018,  0.7251,  0.0845,\n","          -0.6230,  0.5776]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.6493, -0.9999, -1.0000, -0.9995,  0.5191, -0.3621, -0.9878,\n","           0.4121,  0.7224]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4794,  0.4079,  0.3036,  0.1524,  0.3295,  0.3352,  0.0281,\n","           0.0429,  0.5081, -0.5613,  0.5941, -0.0148,  0.1081, -0.3399,\n","          -0.2567, -0.1275,  0.4925, -0.3564,  0.1963,  0.0565,  0.2445,\n","           0.4673,  0.2550,  0.6933,  0.5273, -0.1094,  0.2774, -0.6616,\n","           0.5072,  0.6798, -0.3242, -0.1647, -0.2163,  0.1417, -0.0432,\n","           0.3875,  0.0795,  0.7202, -0.3630, -0.5302,  0.1122,  0.4186,\n","           0.0126, -0.0420, -0.1075,  0.0384, -0.6107,  0.3317,  0.5539,\n","          -0.1450,  0.3499,  0.5623, -0.1401,  0.1740, -0.0519, -0.0557,\n","          -0.1090,  0.4024,  0.5583,  0.0383,  0.2500,  0.4224, -0.0736,\n","           0.4568, -0.0720, -0.4141,  0.5333,  0.1483, -0.3049,  0.6172,\n","          -0.1959,  0.6090, -0.5256, -0.3780, -0.1843,  0.1448, -0.4718,\n","           0.5020,  0.5311, -0.1367,  0.6267, -0.0513, -0.4844, -0.7379,\n","           0.1043, -0.1080, -0.0649,  0.4633, -0.1591,  0.1042, -0.3607,\n","           0.1091, -0.2709,  0.1192,  0.1761, -0.4575, -0.0796, -0.4819,\n","           0.0262, -0.6290, -0.3711, -0.3341, -0.3317, -0.3723, -0.0734,\n","          -0.4956, -0.1326,  0.0279, -0.6572, -0.2405, -0.4810,  0.0468,\n","          -0.3382, -0.8527,  0.1673, -0.2420, -0.1135, -0.4794, -0.4673,\n","           0.5442, -0.6115,  0.1689,  0.7958, -0.5516, -0.5299,  0.4490,\n","          -0.3521, -0.4221]],\n","\n","        [[-0.3766, -0.3372, -0.7113,  0.3573,  0.9254,  0.7064, -0.8117,\n","          -0.8958, -0.9858,  0.3406,  0.9424, -0.2508, -0.4570,  0.4970,\n","           0.9676,  0.1900, -0.3929,  0.9898, -0.8799,  0.6662, -0.9696,\n","          -0.9032, -0.8816, -0.6872, -0.8738,  0.8005, -0.9112,  0.7377,\n","          -0.0447,  0.7080, -0.6547,  0.9123, -0.8575,  0.4335, -0.9778,\n","           0.9368,  0.8191,  0.9540,  0.2098, -0.8935,  0.9477, -0.7866,\n","           0.9582,  0.2136,  0.6460, -0.3516, -0.2450,  0.8536,  0.6180,\n","          -0.6858, -0.2675, -0.7907, -0.2347, -0.7602, -0.8040,  0.9905,\n","           0.5078, -0.9198, -0.8994,  0.9674,  0.9544,  0.7895, -0.7400,\n","           0.9808,  0.5577,  0.8023,  0.5732,  0.9505,  0.9086, -0.9314,\n","          -0.8760,  0.7351, -0.6419,  0.8916,  0.9813,  0.8819, -0.5810,\n","           0.9416,  0.2846,  0.9578,  0.9687, -0.4658, -0.9821, -0.8239,\n","          -0.6077,  0.9562, -0.7274, -0.0822,  0.7957,  0.4106, -0.7384,\n","           0.9948,  0.9233,  0.8556, -0.1011,  0.9110, -0.9916, -0.9070,\n","           0.9171,  0.9866, -0.8416, -0.4709,  0.3781, -0.9138,  0.2095,\n","          -0.9533,  0.4230,  0.9715,  0.8368, -0.0200,  0.9140,  0.1275,\n","           0.9242,  0.8589,  0.9406,  0.5244, -0.9049,  0.4609,  0.6979,\n","          -0.9898,  0.9393,  0.1540,  0.9692, -0.1020,  0.7076, -0.9683,\n","          -0.5535, -0.1042]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.4618, -0.9999, -1.0000, -0.9988,  0.4815, -0.6661, -0.9956,\n","           0.2735,  0.6786]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4761,  0.4616,  0.2132,  0.4040,  0.3635,  0.0966, -0.0899,\n","           0.0065,  0.3514, -0.4707,  0.3391,  0.0543,  0.0343, -0.1905,\n","          -0.2344, -0.1939,  0.4569, -0.3993,  0.2593, -0.0058,  0.3702,\n","           0.4613,  0.2054,  0.5757,  0.4637, -0.2410,  0.2004, -0.3773,\n","           0.3422,  0.5467, -0.2579, -0.2211, -0.2052,  0.2105, -0.0593,\n","           0.3161,  0.0784,  0.8098, -0.3472, -0.4323,  0.1083,  0.2423,\n","           0.2200, -0.0921,  0.0528,  0.1591, -0.3995,  0.0906,  0.2744,\n","          -0.1700,  0.2682,  0.5367, -0.1184,  0.2969,  0.0183, -0.0739,\n","          -0.0297,  0.2930,  0.6072, -0.0442,  0.3672,  0.4571, -0.0538,\n","           0.3176, -0.1348, -0.4083,  0.4210,  0.0042, -0.1581,  0.5815,\n","          -0.2516,  0.7124, -0.4895, -0.2522, -0.1510,  0.1114, -0.5152,\n","           0.2660,  0.5600, -0.1032,  0.4180, -0.1457, -0.5416, -0.5079,\n","           0.1115, -0.2299,  0.0339,  0.2353, -0.0898,  0.1872, -0.4019,\n","           0.0652, -0.5041,  0.1799,  0.1864, -0.4037, -0.0979, -0.3130,\n","           0.0503, -0.4797, -0.2631, -0.0530, -0.2883, -0.2864, -0.0503,\n","          -0.5319,  0.0164,  0.1167, -0.6891, -0.2802, -0.4551, -0.0371,\n","          -0.3569, -0.6212,  0.2428, -0.2462, -0.0660, -0.2946, -0.3659,\n","           0.5654, -0.3968,  0.2463,  0.8367, -0.4992, -0.4126,  0.4216,\n","          -0.1791, -0.3591]],\n","\n","        [[-0.3787, -0.2798, -0.6264,  0.5783,  0.9529,  0.5046, -0.7093,\n","          -0.8939, -0.9080,  0.1857,  0.7493, -0.0055, -0.4699,  0.4901,\n","           0.9605,  0.1262, -0.5169,  0.9767, -0.8616,  0.6751, -0.9908,\n","          -0.8013, -0.8774, -0.6956, -0.8440,  0.8272, -0.8147,  0.7941,\n","          -0.0031,  0.7941, -0.5486,  0.9046, -0.9632,  0.4041, -0.9101,\n","           0.9295,  0.6179,  0.8777,  0.1492, -0.8018,  0.9882, -0.7844,\n","           0.9684,  0.2979,  0.5863, -0.3520, -0.2340,  0.8624,  0.6234,\n","          -0.7120, -0.2852, -0.7308, -0.2288, -0.7590, -0.9537,  0.9345,\n","           0.4180, -0.8632, -0.8090,  0.9721,  0.9656,  0.8910, -0.5251,\n","           0.9908,  0.6286,  0.8714,  0.7554,  0.9381,  0.7243, -0.9854,\n","          -0.9023,  0.6559, -0.6118,  0.9361,  0.7318,  0.8890, -0.6521,\n","           0.7366,  0.1134,  0.9168,  0.5015, -0.4630, -0.9870, -0.8638,\n","          -0.6397,  0.9515, -0.8702, -0.8078,  0.7408,  0.4113, -0.4894,\n","           0.9989,  0.8728,  0.8233, -0.2108,  0.9806, -0.9755, -0.8551,\n","           0.9206,  0.9013, -0.8433, -0.4854,  0.4671, -0.9190,  0.2584,\n","          -0.9316,  0.2890,  0.9146,  0.9129,  0.4968,  0.7812,  0.5411,\n","           0.9086,  0.8681,  0.8465,  0.4753, -0.9452,  0.5597,  0.6168,\n","          -0.9899,  0.9436,  0.4660,  0.6226,  0.0172,  0.7056, -0.9935,\n","          -0.5329, -0.3406]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.3892, -0.9999, -1.0000, -0.9975,  0.4903, -0.7116, -0.9967,\n","           0.2222,  0.6455]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4742,  0.4703,  0.1358,  0.4860,  0.3665, -0.0148, -0.0868,\n","          -0.0188,  0.2568, -0.3772,  0.2303,  0.0920,  0.0406, -0.1230,\n","          -0.2506, -0.2328,  0.4310, -0.3935,  0.2356, -0.0155,  0.4218,\n","           0.4530,  0.1632,  0.4921,  0.4161, -0.2799,  0.1840, -0.1895,\n","           0.2231,  0.4419, -0.2314, -0.2264, -0.1710,  0.2455, -0.0750,\n","           0.2294,  0.0022,  0.8159, -0.3083, -0.3538,  0.0904,  0.1148,\n","           0.2381, -0.1151,  0.1052,  0.2129, -0.2708, -0.0213,  0.1630,\n","          -0.1724,  0.1969,  0.4829, -0.1137,  0.3734,  0.0587, -0.0775,\n","           0.0160,  0.2100,  0.6288, -0.0706,  0.3769,  0.4029, -0.0125,\n","           0.2549, -0.1331, -0.3598,  0.3299, -0.0816, -0.0743,  0.5141,\n","          -0.2011,  0.7211, -0.4629, -0.1818, -0.1459,  0.1111, -0.5247,\n","           0.1709,  0.5832, -0.0968,  0.2690, -0.1812, -0.5151, -0.3482,\n","           0.0801, -0.2679,  0.0612,  0.1263, -0.0370,  0.2112, -0.3796,\n","           0.0999, -0.5265,  0.2360,  0.1961, -0.3964, -0.0550, -0.1960,\n","           0.0227, -0.3571, -0.2380,  0.0817, -0.2476, -0.2568, -0.0512,\n","          -0.5242,  0.0568,  0.1767, -0.6642, -0.2821, -0.4479, -0.0952,\n","          -0.3122, -0.4520,  0.2815, -0.2622, -0.0516, -0.1642, -0.3126,\n","           0.5202, -0.2585,  0.2353,  0.8314, -0.4825, -0.3355,  0.4037,\n","          -0.1195, -0.3348]],\n","\n","        [[-0.3727, -0.2320, -0.6124,  0.6680,  0.9549,  0.4032, -0.6623,\n","          -0.8557, -0.7810,  0.1178,  0.6182,  0.1459, -0.4965,  0.4763,\n","           0.9459,  0.0559, -0.5097,  0.9688, -0.8509,  0.6866, -0.9879,\n","          -0.7062, -0.8517, -0.7062, -0.7979,  0.8494, -0.7042,  0.8131,\n","           0.0349,  0.8515, -0.4440,  0.8832, -0.9514,  0.4062, -0.8353,\n","           0.9131,  0.4933,  0.8171,  0.0869, -0.6928,  0.9920, -0.7796,\n","           0.9572,  0.3667,  0.5096, -0.3528, -0.2313,  0.8639,  0.6305,\n","          -0.7038, -0.3068, -0.6297, -0.2238, -0.7605, -0.9481,  0.9232,\n","           0.3483, -0.8007, -0.7222,  0.9675,  0.9630,  0.9181, -0.4511,\n","           0.9886,  0.6811,  0.8789,  0.8762,  0.9167,  0.6164, -0.9955,\n","          -0.8866,  0.5502, -0.5316,  0.9592,  0.5373,  0.8671, -0.6797,\n","           0.6206, -0.0505,  0.8542,  0.3187, -0.4616, -0.9795, -0.8821,\n","          -0.6602,  0.9378, -0.8937, -0.9034,  0.6834,  0.4129, -0.2977,\n","           0.9976,  0.7875,  0.7962, -0.2661,  0.9745, -0.9523, -0.7507,\n","           0.9186,  0.8439, -0.8450, -0.5287,  0.4639, -0.9051,  0.2344,\n","          -0.8888,  0.1704,  0.8059,  0.9492,  0.5705,  0.6638,  0.7314,\n","           0.8666,  0.8753,  0.8208,  0.4280, -0.9354,  0.5987,  0.5366,\n","          -0.9806,  0.9279,  0.5808,  0.5683,  0.1259,  0.7043, -0.9948,\n","          -0.5218, -0.4073]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.4132, -0.9999, -1.0000, -0.9950,  0.4908, -0.7075, -0.9971,\n","           0.1965,  0.6029]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4736,  0.4567,  0.0883,  0.5044,  0.3526, -0.0658, -0.0584,\n","          -0.0401,  0.2102, -0.2994,  0.2042,  0.1148,  0.0683, -0.1024,\n","          -0.2690, -0.2500,  0.4171, -0.3918,  0.1864, -0.0065,  0.4230,\n","           0.4355,  0.1285,  0.4467,  0.3747, -0.2851,  0.1884, -0.0863,\n","           0.1485,  0.3736, -0.2231, -0.2070, -0.1488,  0.2582, -0.0939,\n","           0.1625, -0.0585,  0.7930, -0.2616, -0.3014,  0.0768,  0.0269,\n","           0.2188, -0.1191,  0.1127,  0.2303, -0.2100, -0.0595,  0.1351,\n","          -0.1661,  0.1404,  0.4282, -0.1169,  0.4004,  0.0786, -0.0710,\n","           0.0495,  0.1568,  0.6345, -0.0673,  0.3621,  0.3407,  0.0189,\n","           0.2372, -0.1092, -0.3154,  0.2695, -0.1078, -0.0311,  0.4664,\n","          -0.1376,  0.6985, -0.4452, -0.1513, -0.1425,  0.1133, -0.5210,\n","           0.1380,  0.5972, -0.1015,  0.1890, -0.1854, -0.4889, -0.2681,\n","           0.0479, -0.2505,  0.0577,  0.0819,  0.0041,  0.2142, -0.3460,\n","           0.1513, -0.4851,  0.2782,  0.2021, -0.4034, -0.0126, -0.1308,\n","          -0.0267, -0.2848, -0.2393,  0.1306, -0.2234, -0.2634, -0.0618,\n","          -0.5086,  0.0665,  0.2160, -0.6450, -0.2666, -0.4496, -0.1190,\n","          -0.2606, -0.3788,  0.3051, -0.2808, -0.0483, -0.0878, -0.2916,\n","           0.4643, -0.1921,  0.2076,  0.8016, -0.4812, -0.2924,  0.3908,\n","          -0.1053, -0.3201]],\n","\n","        [[-0.3611, -0.1817, -0.6270,  0.6966,  0.9457,  0.3548, -0.5858,\n","          -0.8068, -0.6403,  0.0929,  0.5121,  0.2293, -0.5234,  0.4666,\n","           0.9302, -0.0073, -0.4356,  0.9596, -0.8436,  0.7006, -0.9766,\n","          -0.6283, -0.8201, -0.7161, -0.7368,  0.8624, -0.5908,  0.8037,\n","           0.0679,  0.8890, -0.3475,  0.8489, -0.9183,  0.4143, -0.7560,\n","           0.8998,  0.3988,  0.7616,  0.0242, -0.5821,  0.9915, -0.7710,\n","           0.9461,  0.4227,  0.4319, -0.3539, -0.2311,  0.8565,  0.6391,\n","          -0.6956, -0.3280, -0.5017, -0.2186, -0.7628, -0.9270,  0.9229,\n","           0.2960, -0.7379, -0.6342,  0.9541,  0.9530,  0.9145, -0.4341,\n","           0.9852,  0.7197,  0.8755,  0.9363,  0.8845,  0.5426, -0.9970,\n","          -0.8724,  0.4348, -0.4244,  0.9688,  0.3744,  0.8441, -0.6903,\n","           0.5250, -0.1795,  0.7885,  0.2289, -0.4612, -0.9707, -0.8878,\n","          -0.6764,  0.9208, -0.8890, -0.9154,  0.6301,  0.4161, -0.1618,\n","           0.9958,  0.7117,  0.7753, -0.2835,  0.9561, -0.9215, -0.6312,\n","           0.9127,  0.8032, -0.8470, -0.5778,  0.4242, -0.8792,  0.2018,\n","          -0.8270,  0.0716,  0.7064,  0.9644,  0.5905,  0.5767,  0.8248,\n","           0.8170,  0.8797,  0.7872,  0.3826, -0.8988,  0.5908,  0.4682,\n","          -0.9683,  0.9085,  0.6224,  0.5348,  0.2198,  0.7036, -0.9930,\n","          -0.5193, -0.4125]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.4823, -0.9999, -1.0000, -0.9904,  0.4850, -0.6850, -0.9973,\n","           0.1881,  0.5555]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4747,  0.4393,  0.0699,  0.4995,  0.3345, -0.0829, -0.0338,\n","          -0.0595,  0.1937, -0.2461,  0.2137,  0.1277,  0.0893, -0.1018,\n","          -0.2820, -0.2522,  0.4123, -0.3992,  0.1438,  0.0031,  0.4080,\n","           0.4151,  0.1017,  0.4283,  0.3438, -0.2790,  0.1977, -0.0406,\n","           0.1063,  0.3372, -0.2225, -0.1848, -0.1414,  0.2585, -0.1138,\n","           0.1245, -0.0867,  0.7643, -0.2246, -0.2706,  0.0704, -0.0289,\n","           0.2002, -0.1138,  0.1048,  0.2308, -0.1875, -0.0578,  0.1437,\n","          -0.1587,  0.1012,  0.3875, -0.1213,  0.4016,  0.0850, -0.0623,\n","           0.0734,  0.1310,  0.6336, -0.0527,  0.3473,  0.2973,  0.0358,\n","           0.2412, -0.0863, -0.2901,  0.2365, -0.1029, -0.0139,  0.4412,\n","          -0.0933,  0.6700, -0.4336, -0.1416, -0.1376,  0.1100, -0.5161,\n","           0.1302,  0.6016, -0.1078,  0.1569, -0.1791, -0.4756, -0.2417,\n","           0.0304, -0.2192,  0.0462,  0.0719,  0.0300,  0.2127, -0.3222,\n","           0.1947, -0.4395,  0.3018,  0.2046, -0.4122,  0.0152, -0.0990,\n","          -0.0729, -0.2586, -0.2460,  0.1349, -0.2131, -0.2809, -0.0724,\n","          -0.4958,  0.0677,  0.2392, -0.6347, -0.2474, -0.4522, -0.1197,\n","          -0.2268, -0.3649,  0.3208, -0.2934, -0.0477, -0.0521, -0.2876,\n","           0.4189, -0.1722,  0.1863,  0.7673, -0.4839, -0.2740,  0.3842,\n","          -0.1081, -0.3089]],\n","\n","        [[-0.3465, -0.1280, -0.6512,  0.6990,  0.9315,  0.3326, -0.5045,\n","          -0.7589, -0.5184,  0.0828,  0.4345,  0.2679, -0.5439,  0.4609,\n","           0.9143, -0.0596, -0.3392,  0.9515, -0.8382,  0.7166, -0.9582,\n","          -0.5694, -0.7869, -0.7252, -0.6686,  0.8657, -0.4835,  0.7769,\n","           0.0963,  0.9126, -0.2655,  0.8083, -0.8774,  0.4198, -0.6804,\n","           0.8931,  0.3239,  0.7204, -0.0374, -0.4798,  0.9907, -0.7577,\n","           0.9366,  0.4666,  0.3631, -0.3552, -0.2304,  0.8398,  0.6485,\n","          -0.6949, -0.3481, -0.3708, -0.2131, -0.7651, -0.9034,  0.9274,\n","           0.2578, -0.6799, -0.5482,  0.9362,  0.9375,  0.8981, -0.4402,\n","           0.9814,  0.7469,  0.8711,  0.9550,  0.8425,  0.4920, -0.9963,\n","          -0.8644,  0.3236, -0.3165,  0.9719,  0.2580,  0.8282, -0.6939,\n","           0.4690, -0.2690,  0.7237,  0.1941, -0.4617, -0.9624, -0.8875,\n","          -0.6906,  0.9060, -0.8782, -0.9115,  0.5933,  0.4209, -0.0830,\n","           0.9936,  0.6528,  0.7592, -0.2765,  0.9298, -0.8874, -0.5215,\n","           0.9061,  0.7821, -0.8493, -0.6234,  0.3652, -0.8484,  0.1802,\n","          -0.7535, -0.0100,  0.6380,  0.9700,  0.5982,  0.5213,  0.8697,\n","           0.7746,  0.8805,  0.7545,  0.3418, -0.8506,  0.5624,  0.4125,\n","          -0.9557,  0.8877,  0.6346,  0.5349,  0.2967,  0.7031, -0.9905,\n","          -0.5210, -0.3858]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.5482, -0.9999, -1.0000, -0.9833,  0.4842, -0.6585, -0.9974,\n","           0.1975,  0.5159]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4767,  0.4269,  0.0691,  0.4884,  0.3183, -0.0828, -0.0206,\n","          -0.0763,  0.1913, -0.2148,  0.2321,  0.1337,  0.0978, -0.1057,\n","          -0.2891, -0.2469,  0.4124, -0.4107,  0.1176,  0.0077,  0.3951,\n","           0.3984,  0.0822,  0.4234,  0.3242, -0.2705,  0.2057, -0.0254,\n","           0.0837,  0.3215, -0.2241, -0.1692, -0.1439,  0.2544, -0.1317,\n","           0.1085, -0.0924,  0.7413, -0.2025, -0.2545,  0.0697, -0.0620,\n","           0.1901, -0.1065,  0.0943,  0.2247, -0.1811, -0.0423,  0.1614,\n","          -0.1537,  0.0771,  0.3643, -0.1240,  0.3937,  0.0847, -0.0558,\n","           0.0888,  0.1234,  0.6314, -0.0380,  0.3379,  0.2771,  0.0422,\n","           0.2505, -0.0713, -0.2801,  0.2216, -0.0876, -0.0120,  0.4296,\n","          -0.0711,  0.6471, -0.4258, -0.1401, -0.1319,  0.1025, -0.5138,\n","           0.1295,  0.5998, -0.1120,  0.1507, -0.1732, -0.4718, -0.2418,\n","           0.0269, -0.1968,  0.0368,  0.0762,  0.0423,  0.2122, -0.3114,\n","           0.2229, -0.4090,  0.3101,  0.2055, -0.4178,  0.0294, -0.0851,\n","          -0.1039, -0.2578, -0.2508,  0.1226, -0.2105, -0.2964, -0.0789,\n","          -0.4885,  0.0665,  0.2512, -0.6301, -0.2308, -0.4524, -0.1113,\n","          -0.2119, -0.3750,  0.3311, -0.2986, -0.0470, -0.0407, -0.2902,\n","           0.3893, -0.1750,  0.1740,  0.7414, -0.4850, -0.2700,  0.3828,\n","          -0.1148, -0.3003]],\n","\n","        [[-0.3306, -0.0724, -0.6744,  0.6935,  0.9159,  0.3217, -0.4370,\n","          -0.7190, -0.4318,  0.0804,  0.3861,  0.2797, -0.5566,  0.4577,\n","           0.8994, -0.1007, -0.2517,  0.9466, -0.8339,  0.7333, -0.9359,\n","          -0.5272, -0.7557, -0.7336, -0.6035,  0.8618, -0.3871,  0.7438,\n","           0.1212,  0.9269, -0.2021,  0.7678, -0.8407,  0.4194, -0.6167,\n","           0.8920,  0.2632,  0.6952, -0.0965, -0.3917,  0.9901, -0.7399,\n","           0.9305,  0.5007,  0.3067, -0.3567, -0.2282,  0.8158,  0.6583,\n","          -0.7004, -0.3667, -0.2554, -0.2069, -0.7671, -0.8824,  0.9334,\n","           0.2296, -0.6297, -0.4683,  0.9191,  0.9195,  0.8791, -0.4519,\n","           0.9776,  0.7646,  0.8664,  0.9541,  0.7942,  0.4611, -0.9952,\n","          -0.8617,  0.2283, -0.2290,  0.9718,  0.1864,  0.8202, -0.6943,\n","           0.4509, -0.3260,  0.6704,  0.1880, -0.4624, -0.9558, -0.8850,\n","          -0.7029,  0.8952, -0.8672, -0.9017,  0.5741,  0.4273, -0.0459,\n","           0.9915,  0.6116,  0.7457, -0.2591,  0.9004, -0.8552, -0.4403,\n","           0.9013,  0.7757, -0.8516, -0.6623,  0.3024, -0.8194,  0.1762,\n","          -0.6775, -0.0778,  0.6050,  0.9715,  0.6026,  0.4897,  0.8872,\n","           0.7464,  0.8769,  0.7281,  0.3067, -0.8051,  0.5334,  0.3680,\n","          -0.9443,  0.8683,  0.6364,  0.5527,  0.3560,  0.7027, -0.9880,\n","          -0.5232, -0.3435]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[-0.5885, -0.9998, -1.0000, -0.9730,  0.4941, -0.6393, -0.9975,\n","           0.2219,  0.4913]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.4785,  0.4205,  0.0747,  0.4786,  0.3060, -0.0765, -0.0168,\n","          -0.0892,  0.1929, -0.1987,  0.2471,  0.1354,  0.0975, -0.1082,\n","          -0.2922, -0.2397,  0.4140, -0.4215,  0.1057,  0.0081,  0.3889,\n","           0.3878,  0.0687,  0.4229,  0.3135, -0.2634,  0.2108, -0.0222,\n","           0.0719,  0.3161, -0.2256, -0.1611, -0.1503,  0.2503, -0.1460,\n","           0.1043, -0.0878,  0.7273, -0.1926, -0.2470,  0.0716, -0.0806,\n","           0.1870, -0.1009,  0.0866,  0.2173, -0.1793, -0.0280,  0.1757,\n","          -0.1516,  0.0635,  0.3545, -0.1248,  0.3855,  0.0827, -0.0525,\n","           0.0972,  0.1249,  0.6298, -0.0277,  0.3330,  0.2731,  0.0433,\n","           0.2572, -0.0638, -0.2781,  0.2165, -0.0731, -0.0164,  0.4242,\n","          -0.0637,  0.6327, -0.4201, -0.1404, -0.1269,  0.0945, -0.5144,\n","           0.1287,  0.5960, -0.1135,  0.1547, -0.1701, -0.4722, -0.2494,\n","           0.0306, -0.1873,  0.0321,  0.0830,  0.0462,  0.2135, -0.3098,\n","           0.2374, -0.3942,  0.3102,  0.2061, -0.4195,  0.0348, -0.0794,\n","          -0.1195, -0.2647, -0.2522,  0.1086, -0.2103, -0.3057, -0.0815,\n","          -0.4857,  0.0647,  0.2569, -0.6282, -0.2187, -0.4504, -0.1026,\n","          -0.2086, -0.3882,  0.3374, -0.2996, -0.0461, -0.0402, -0.2941,\n","           0.3738, -0.1841,  0.1677,  0.7266, -0.4830, -0.2720,  0.3844,\n","          -0.1198, -0.2937]],\n","\n","        [[-0.3143, -0.0175, -0.6924,  0.6888,  0.9010,  0.3152, -0.3889,\n","          -0.6889, -0.3776,  0.0845,  0.3594,  0.2774, -0.5628,  0.4556,\n","           0.8868, -0.1326, -0.1858,  0.9451, -0.8303,  0.7491, -0.9141,\n","          -0.4967, -0.7290, -0.7414, -0.5472,  0.8544, -0.3028,  0.7120,\n","           0.1437,  0.9356, -0.1567,  0.7303, -0.8135,  0.4128, -0.5680,\n","           0.8936,  0.2126,  0.6811, -0.1517, -0.3194,  0.9895, -0.7186,\n","           0.9281,  0.5273,  0.2614, -0.3582, -0.2246,  0.7881,  0.6680,\n","          -0.7087, -0.3833, -0.1617, -0.2002, -0.7687, -0.8670,  0.9387,\n","           0.2078, -0.5874, -0.3979,  0.9056,  0.9023,  0.8628, -0.4614,\n","           0.9746,  0.7742,  0.8613,  0.9483,  0.7433,  0.4447, -0.9942,\n","          -0.8622,  0.1535, -0.1686,  0.9704,  0.1449,  0.8181, -0.6933,\n","           0.4522, -0.3616,  0.6324,  0.1906, -0.4630, -0.9511, -0.8826,\n","          -0.7132,  0.8881, -0.8571, -0.8886,  0.5667,  0.4346, -0.0322,\n","           0.9900,  0.5845,  0.7332, -0.2414,  0.8731, -0.8274, -0.3918,\n","           0.8988,  0.7759, -0.8537, -0.6941,  0.2458, -0.7958,  0.1880,\n","          -0.6060, -0.1350,  0.5953,  0.9714,  0.6070,  0.4720,  0.8882,\n","           0.7309,  0.8688,  0.7098,  0.2771, -0.7708,  0.5127,  0.3328,\n","          -0.9343,  0.8523,  0.6360,  0.5720,  0.3997,  0.7021, -0.9861,\n","          -0.5240, -0.2967]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[ 0.3890, -0.9996, -1.0000, -0.9469,  0.5605, -0.8491, -0.9982,\n","           0.1787,  0.4750]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.3504,  0.3332,  0.0416,  0.5189,  0.3242,  0.0245,  0.0709,\n","          -0.0323,  0.1293, -0.4057,  0.2539,  0.0678,  0.0016, -0.1003,\n","          -0.2829, -0.1481,  0.4246, -0.3351,  0.1319,  0.0360,  0.4472,\n","           0.2966, -0.0500,  0.3607,  0.3187, -0.1698,  0.1403,  0.0215,\n","           0.1088,  0.1667, -0.2390, -0.3027, -0.2305,  0.1748, -0.0029,\n","           0.2902,  0.1995,  0.6756, -0.3067, -0.2899,  0.1617, -0.0278,\n","           0.2556, -0.1444,  0.1236,  0.2184, -0.2855, -0.0734,  0.1659,\n","          -0.1065, -0.0654,  0.2530, -0.1514,  0.2946, -0.0094, -0.2038,\n","          -0.2426,  0.3155,  0.5376, -0.1856,  0.4144,  0.3892,  0.0101,\n","           0.1661, -0.2854, -0.5452,  0.3295, -0.0278, -0.1281,  0.3955,\n","          -0.2592,  0.5580, -0.2560,  0.0543, -0.0383,  0.0125, -0.3337,\n","           0.1713,  0.4646, -0.1538,  0.0888, -0.3080, -0.5369, -0.2661,\n","          -0.0715, -0.3059,  0.1471,  0.1277, -0.1563,  0.1705, -0.3283,\n","           0.2249, -0.3681,  0.0942,  0.0677, -0.2485, -0.2927, -0.1530,\n","          -0.1065, -0.1714, -0.1657,  0.0445, -0.1699, -0.2106, -0.0110,\n","          -0.4630, -0.0219,  0.0014, -0.4997, -0.3611, -0.2292, -0.0984,\n","          -0.3317, -0.4763,  0.2464, -0.0554,  0.1802, -0.1563, -0.3853,\n","           0.3681, -0.2514,  0.3258,  0.5765, -0.2783, -0.2017,  0.2811,\n","          -0.1184, -0.1977]],\n","\n","        [[-0.3373,  0.0136, -0.6522,  0.6718,  0.9025,  0.2536, -0.4306,\n","          -0.6852, -0.4358, -0.0214,  0.4060,  0.3438, -0.3864,  0.3457,\n","           0.8450, -0.1853, -0.2901,  0.9204, -0.7816,  0.7400, -0.9493,\n","          -0.4248, -0.7852, -0.7441, -0.5644,  0.8654, -0.1898,  0.7440,\n","           0.1791,  0.9194, -0.2315,  0.6982, -0.8801,  0.4181, -0.5983,\n","           0.8359,  0.1670,  0.7416, -0.1988, -0.2172,  0.9793, -0.6913,\n","           0.9153,  0.5497,  0.2225, -0.3439, -0.1800,  0.7918,  0.6622,\n","          -0.6154, -0.3662, -0.0746, -0.1791, -0.7307, -0.8351,  0.8771,\n","           0.1397, -0.5193, -0.3160,  0.9070,  0.8928,  0.8786, -0.1575,\n","           0.9535,  0.7322,  0.7946,  0.9416,  0.6964,  0.5163, -0.9927,\n","          -0.8414,  0.1896, -0.2318,  0.9501,  0.2370,  0.8179, -0.6023,\n","           0.6771, -0.4333,  0.7210,  0.2266, -0.4471, -0.9497, -0.8170,\n","          -0.6691,  0.8397, -0.8358, -0.8546,  0.5879,  0.4324, -0.2978,\n","           0.9898,  0.6441,  0.6640, -0.3170,  0.9126, -0.8233, -0.4409,\n","           0.8674,  0.7561, -0.8176, -0.6624,  0.2656, -0.8090,  0.2316,\n","          -0.5977, -0.2058,  0.6646,  0.9529,  0.4525,  0.3740,  0.8819,\n","           0.7592,  0.8494,  0.7529,  0.2290, -0.8205,  0.5180,  0.2691,\n","          -0.9342,  0.8574,  0.6238,  0.5429,  0.4552,  0.6642, -0.9772,\n","          -0.4514, -0.3545]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[ 0.8434, -0.9992, -1.0000, -0.8899,  0.5156, -0.9088, -0.9987,\n","           0.0252,  0.2939]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[-0.2046,  0.2745,  0.0260,  0.5035,  0.2935,  0.0379,  0.0729,\n","           0.0129,  0.0486, -0.4805,  0.0836,  0.0523, -0.0463, -0.0627,\n","          -0.2142, -0.1029,  0.3852, -0.2116,  0.1691,  0.0114,  0.4692,\n","           0.2494, -0.1062,  0.2259,  0.3126, -0.1059,  0.0461,  0.1253,\n","           0.1341,  0.0376, -0.2032, -0.3634, -0.1864,  0.1629,  0.0912,\n","           0.3347,  0.3138,  0.5947, -0.3459, -0.2893,  0.1676, -0.0168,\n","           0.2713, -0.1681,  0.1786,  0.2208, -0.2719, -0.1788,  0.0743,\n","          -0.1059, -0.1145,  0.1637, -0.1534,  0.2527, -0.0579, -0.2742,\n","          -0.3705,  0.3884,  0.4448, -0.2510,  0.3911,  0.3964, -0.0060,\n","           0.0012, -0.4161, -0.5766,  0.3383, -0.0354, -0.1646,  0.2937,\n","          -0.3536,  0.4920, -0.1131,  0.1857,  0.0299,  0.0043, -0.2119,\n","           0.1415,  0.4028, -0.1206,  0.0175, -0.3467, -0.4392, -0.1386,\n","          -0.1340, -0.3705,  0.2355,  0.0794, -0.2095,  0.1405, -0.2940,\n","           0.1040, -0.3569, -0.0177,  0.0050, -0.0520, -0.3750, -0.1512,\n","          -0.0426, -0.0819, -0.0935,  0.0841, -0.1190, -0.0736,  0.0869,\n","          -0.3709, -0.0569, -0.1306, -0.3334, -0.4151, -0.0632, -0.1261,\n","          -0.3640, -0.4012,  0.1567,  0.0417,  0.2457, -0.1536, -0.4020,\n","           0.3822, -0.1905,  0.3345,  0.4432, -0.1267, -0.1105,  0.1847,\n","          -0.0800, -0.0992]],\n","\n","        [[-0.3527,  0.0269, -0.5555,  0.6498,  0.8912,  0.2154, -0.4849,\n","          -0.6541, -0.4597, -0.1223,  0.3839,  0.4394, -0.1833,  0.2093,\n","           0.7950, -0.2320, -0.3909,  0.8516, -0.6978,  0.7273, -0.9685,\n","          -0.3277, -0.8121, -0.7338, -0.5709,  0.8713, -0.0819,  0.7742,\n","           0.2154,  0.8904, -0.2848,  0.6431, -0.8931,  0.4518, -0.6070,\n","           0.7007,  0.1071,  0.7475, -0.2368, -0.1120,  0.9393, -0.6623,\n","           0.8729,  0.5690,  0.1807, -0.3155, -0.1323,  0.7960,  0.6459,\n","          -0.4309, -0.3285,  0.0051, -0.1515, -0.6568, -0.7851,  0.7519,\n","           0.0602, -0.4342, -0.2377,  0.8927,  0.8935,  0.8780,  0.1893,\n","           0.8944,  0.6451,  0.7018,  0.9308,  0.6509,  0.5393, -0.9881,\n","          -0.7670,  0.2276, -0.2931,  0.9211,  0.1885,  0.7755, -0.4462,\n","           0.7197, -0.4908,  0.7544,  0.1337, -0.4237, -0.9247, -0.7346,\n","          -0.5723,  0.7286, -0.7606, -0.8321,  0.5405,  0.4261, -0.4161,\n","           0.9831,  0.6574,  0.5500, -0.3812,  0.9329, -0.8030, -0.4705,\n","           0.7908,  0.6687, -0.7125, -0.5991,  0.3054, -0.8129,  0.2337,\n","          -0.5959, -0.2658,  0.5755,  0.9238,  0.2558,  0.2169,  0.8852,\n","           0.7315,  0.8242,  0.7617,  0.1797, -0.8405,  0.5093,  0.1954,\n","          -0.8918,  0.8506,  0.5926,  0.3196,  0.5033,  0.5945, -0.9517,\n","          -0.3392, -0.4487]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[ 0.9914, -0.9997, -1.0000, -0.9426, -0.8811, -0.9867, -0.9999,\n","          -0.9784, -0.9576]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[ 2.1816e-01, -2.3248e-01, -2.4495e-01,  2.6332e-01, -5.2104e-02,\n","          -2.5980e-01,  1.0990e-01,  1.9752e-01, -1.4741e-01, -5.7686e-01,\n","          -7.7274e-02,  9.8042e-02, -2.7132e-01,  1.9431e-01,  2.7142e-01,\n","           1.9482e-01,  2.8319e-01,  2.6920e-01,  3.8546e-01, -1.2635e-01,\n","           4.6566e-01, -1.1772e-01, -4.1092e-01, -1.4647e-01,  1.7705e-03,\n","           3.2706e-01, -3.6324e-01,  1.8786e-01,  2.3012e-01, -3.8143e-02,\n","          -4.5602e-02, -3.9674e-01,  2.1799e-01, -1.1054e-01,  3.5610e-01,\n","           5.0503e-01,  3.2320e-01,  7.7649e-02, -3.8013e-01, -4.9015e-02,\n","          -1.4392e-01, -6.1702e-02,  6.5213e-02, -7.4350e-02,  1.4575e-01,\n","          -3.1498e-02,  1.3615e-01, -1.4691e-01, -7.2720e-02,  5.1048e-02,\n","          -5.8309e-02, -2.5505e-01, -6.6929e-02,  2.4499e-01, -4.9161e-01,\n","          -5.5370e-01, -5.4329e-01,  3.1170e-01,  2.5166e-01, -6.0275e-01,\n","          -4.3137e-03,  3.4409e-01, -5.0284e-02, -2.2937e-01, -5.3936e-01,\n","          -7.3613e-01,  1.9896e-01,  2.1603e-01, -1.6416e-01,  5.5544e-03,\n","          -2.9205e-01, -2.3589e-01,  1.5090e-01,  4.0898e-01,  5.1042e-01,\n","          -2.3999e-01,  1.4723e-01,  2.4044e-01,  8.2559e-02,  2.6056e-02,\n","           1.2809e-01, -5.1094e-01, -5.0747e-02,  9.4137e-02, -3.0599e-01,\n","          -5.9139e-01,  1.8379e-01,  1.2498e-01, -2.4540e-01,  1.7535e-01,\n","          -1.0250e-01, -4.1104e-01, -3.1219e-01, -2.1377e-01, -3.0999e-01,\n","           8.0827e-02, -6.0855e-01, -1.8252e-01,  2.9209e-01, -2.7491e-02,\n","           2.1809e-01, -1.2456e-01,  1.1327e-01,  2.4359e-02,  1.6317e-01,\n","           6.5330e-02, -2.8662e-02, -4.7046e-01,  2.3983e-01, -2.8567e-01,\n","           3.4290e-01,  1.9648e-01, -4.2766e-01, -4.2699e-01, -2.6781e-01,\n","           3.3493e-01,  5.5974e-01, -2.1946e-01, -3.3754e-01,  5.1242e-01,\n","          -2.7500e-01,  3.1101e-01, -2.9980e-01,  3.8851e-01,  1.0463e-01,\n","          -1.6391e-01,  2.2939e-01,  1.1231e-04]],\n","\n","        [[-4.1085e-01, -8.3884e-03, -1.7130e-01,  4.0338e-01,  9.0804e-01,\n","           1.4556e-01, -6.3898e-01, -6.3572e-01, -5.3041e-01, -4.2849e-01,\n","           4.8598e-01,  6.0585e-01,  4.5360e-01, -2.3935e-01,  4.0393e-01,\n","          -2.9232e-01, -5.5987e-01,  6.6415e-04, -3.3547e-01,  7.0410e-01,\n","          -9.8591e-01, -2.2641e-01, -8.7896e-01, -6.6859e-01, -6.0173e-01,\n","           8.7278e-01,  2.8783e-02,  8.4462e-01,  2.7231e-01,  4.0619e-01,\n","          -4.4046e-01,  5.7537e-01, -9.2081e-01,  5.6806e-01, -6.4489e-01,\n","          -1.8772e-01,  1.7419e-01,  7.9275e-01, -1.2538e-01, -7.4678e-03,\n","           4.8232e-01, -6.0571e-01,  1.5806e-01,  5.9007e-01,  1.5144e-01,\n","          -1.5229e-01,  2.8505e-02,  8.0829e-01,  4.7233e-01,  1.6144e-01,\n","          -5.1015e-02,  6.1890e-02, -3.9588e-02, -3.1581e-01, -3.0322e-01,\n","          -3.3022e-01, -9.0517e-02, -3.1409e-01, -1.5721e-01,  8.5837e-01,\n","           8.3601e-01,  8.1237e-01,  6.3512e-01, -1.6458e-01,  3.3185e-01,\n","           3.0594e-01,  9.2197e-01,  6.2851e-01,  5.0826e-01, -7.7100e-01,\n","          -1.7715e-01,  3.2015e-01, -4.0833e-01,  3.2004e-01,  3.1725e-01,\n","           3.8743e-01,  3.8452e-01,  8.2857e-01, -5.1281e-01,  8.6455e-01,\n","           1.2375e-01, -2.8736e-01, -7.9057e-01, -3.0376e-01, -7.9149e-02,\n","           4.0496e-01, -5.7755e-01, -7.8412e-01,  4.8848e-01,  3.9648e-01,\n","          -7.2724e-01,  8.9788e-01,  6.5597e-01,  2.5453e-01, -4.0462e-01,\n","           9.2756e-01, -8.0020e-01, -4.4527e-01,  3.3260e-01,  3.8296e-01,\n","           1.1606e-01, -3.9790e-01,  4.0047e-01, -8.2221e-01,  3.6249e-02,\n","          -6.4503e-01, -2.9339e-01,  4.6466e-01,  7.3438e-01, -2.9760e-01,\n","          -1.0618e-01,  9.1355e-01,  6.4395e-01,  8.1126e-01,  5.6499e-01,\n","           1.2352e-01, -8.8467e-01,  2.7514e-01,  8.3049e-02, -8.0693e-01,\n","           8.4247e-01,  2.4817e-01, -1.6876e-01,  5.4646e-01,  1.5416e-01,\n","          -3.4268e-01,  3.1672e-02, -6.4926e-01]]], device='cuda:0',\n","       grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","output_logits:  tensor([[[ 0.9996, -0.9996, -1.0000, -0.9032, -0.9987, -0.9999, -1.0000,\n","          -0.9999, -0.9996]]], device='cuda:0', grad_fn=<TanhBackward0>) torch.Size([1, 1, 9])\n","hidden_state:  tensor([[[ 0.5292, -0.5417, -0.3706,  0.0359, -0.2573, -0.4575,  0.1167,\n","           0.3669, -0.3321, -0.5721, -0.2730,  0.0865, -0.3754,  0.3889,\n","           0.6448,  0.3339,  0.1767,  0.5943,  0.4942, -0.2801,  0.4016,\n","          -0.3671, -0.5759, -0.4567, -0.2398,  0.5413, -0.6332,  0.2697,\n","           0.3305, -0.0931,  0.0603, -0.3973,  0.4279, -0.2518,  0.5924,\n","           0.4917,  0.2805, -0.3709, -0.3571,  0.1020, -0.2976, -0.1262,\n","          -0.0920,  0.0670,  0.0090, -0.2725,  0.5767, -0.1350, -0.1935,\n","           0.1481,  0.0781, -0.4634, -0.0119,  0.1430, -0.6825, -0.6770,\n","          -0.5048,  0.2781,  0.0353, -0.7119, -0.2778,  0.2772, -0.0673,\n","          -0.4210, -0.5421, -0.7390,  0.0871,  0.3407, -0.1537, -0.2787,\n","          -0.1950, -0.6370,  0.3827,  0.5349,  0.7434, -0.3332,  0.4591,\n","           0.2063, -0.3200,  0.2090,  0.1276, -0.5238,  0.2836,  0.2827,\n","          -0.3877, -0.6617,  0.1467,  0.0422, -0.1905,  0.1241,  0.0354,\n","          -0.6994, -0.2647, -0.3626, -0.4848,  0.2367, -0.6449, -0.1621,\n","           0.4886,  0.0608,  0.4186, -0.1840,  0.3493,  0.1229,  0.2441,\n","           0.4319, -0.0414, -0.6538,  0.5585, -0.1478,  0.6091,  0.4315,\n","          -0.4283, -0.3152, -0.5146,  0.4594,  0.6458, -0.1872, -0.2118,\n","           0.5538, -0.2335,  0.2591, -0.5855,  0.6418,  0.3123, -0.4752,\n","           0.4448,  0.1702]],\n","\n","        [[-0.4856, -0.0986,  0.2712,  0.0115,  0.8918,  0.0724, -0.7740,\n","          -0.6148, -0.5909, -0.6639,  0.5230,  0.7266,  0.8043, -0.5617,\n","           0.0834, -0.3610, -0.6928, -0.4229,  0.2783,  0.6793, -0.9869,\n","          -0.1101, -0.8866, -0.5378, -0.6351,  0.8699,  0.1415,  0.9058,\n","           0.3780, -0.2914, -0.5891,  0.4821, -0.8692,  0.6999, -0.6696,\n","          -0.6128,  0.2802,  0.8104,  0.1186,  0.0892,  0.0870, -0.4900,\n","          -0.5130,  0.5389,  0.1279,  0.2144,  0.2292,  0.8274,  0.0676,\n","           0.6442,  0.4111,  0.1188,  0.2721,  0.2124,  0.0556, -0.8329,\n","          -0.2745, -0.1697, -0.0720,  0.8010,  0.7377,  0.6579,  0.8373,\n","          -0.6659, -0.0335, -0.0614,  0.9192,  0.6082,  0.4120, -0.3119,\n","           0.3868,  0.4262, -0.5095, -0.6068,  0.3746, -0.1389,  0.8332,\n","           0.8244, -0.5219,  0.9108,  0.0945,  0.1413, -0.6346, -0.0847,\n","           0.4891,  0.1372, -0.3114, -0.7780,  0.3421,  0.3106, -0.8589,\n","           0.4417,  0.6647, -0.2649, -0.4128,  0.8765, -0.7851, -0.3913,\n","          -0.1486,  0.0511,  0.7868,  0.0018,  0.5005, -0.8213, -0.2249,\n","          -0.7381, -0.1938,  0.3070,  0.5472, -0.5597, -0.4869,  0.9293,\n","           0.4436,  0.7961,  0.4291,  0.0502, -0.9160, -0.0227, -0.0754,\n","          -0.5898,  0.8301, -0.1573, -0.5423,  0.5773, -0.5275,  0.2603,\n","           0.4674, -0.8013]]], device='cuda:0', grad_fn=<CudnnRnnBackward0>) torch.Size([2, 1, 128])\n","\n","\n","torch.return_types.topk(\n","values=tensor([[[0.9981],\n","         [1.0000],\n","         [0.7224],\n","         [0.6786],\n","         [0.6455],\n","         [0.6029],\n","         [0.5555],\n","         [0.5159],\n","         [0.4941],\n","         [0.5605],\n","         [0.8434],\n","         [0.9914],\n","         [0.9996]]], device='cuda:0', grad_fn=<TopkBackward0>),\n","indices=tensor([[[5],\n","         [5],\n","         [8],\n","         [8],\n","         [8],\n","         [8],\n","         [8],\n","         [8],\n","         [4],\n","         [4],\n","         [0],\n","         [0],\n","         [0]]], device='cuda:0'))\n"]}],"source":["# Sanity check for trained model output, to compare with the model deployed to the mobile app\n","_, t_out = test_batch(\n","    encoder, \n","    decoder, \n","    torch.zeros(1, 1, 138, 80, dtype=test_spectrograms.dtype).to(device), \n","    torch.tensor([2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=test_input_tokens.dtype).reshape((1, -1)).to(device),\n","    log=True\n",")\n","\n","print(t_out.topk(1))"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"Vn2hzvSYMcFk"},"outputs":[],"source":["# Maps integer tokens back to the syllables they represent\n","tokens_to_syllables = {token:tag for (tag, token) in syl_to_tokens.items()}"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"mn5l3tUjLNd4"},"outputs":[],"source":["# Get predictions for training and test data\n","def training_predictions(start_idx, predictions=5):\n","    _, test_output = test_batch(encoder, decoder, train_spectrograms[start_idx:start_idx + predictions], train_output_tokens[start_idx:start_idx + predictions])\n","    test_predictions = torch.argmax(test_output, dim=-1).cpu().numpy()\n","    return [[tokens_to_syllables[token] for token in prediction] for prediction in test_predictions]\n","\n","def test_predictions(start_idx, predictions=5):\n","    _, test_output = test_batch(encoder, decoder, test_spectrograms[start_idx:start_idx + predictions], train_output_tokens[start_idx:start_idx + predictions])\n","    test_predictions = torch.argmax(test_output, dim=-1).cpu().numpy()\n","    return [[tokens_to_syllables[token] for token in prediction] for prediction in test_predictions]"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":472,"status":"ok","timestamp":1673625200082,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"gPLAxbwJNPYG","outputId":"f0f2cdfd-c913-4f97-b9c8-3035449a91cd"},"outputs":[{"name":"stdout","output_type":"stream","text":["['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['BKG', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['gai4', 'ge1', 'qing3', 'da2', 'hui2', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['BKG', 'END', 'END', 'END', 'END', 'END']\n","\n","\n","\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['BKG', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['gai4', 'gai4', 'qing3', 'qing3', 'qing3', 'qing3']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['BKG', 'END', 'END', 'END', 'END', 'END']\n"]}],"source":["# Compare the ground truth labels with the predicted labels for a sample of the training set\n","ground_truth = [[tokens_to_syllables[token.item()] for token in prediction] for prediction in train_output_tokens[0:10]]\n","for tag in ground_truth:\n","    print(tag)\n","\n","print(\"\\n\\n\")    \n","predictions = training_predictions(0, 10)\n","for tag in predictions:\n","    print(tag)"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":519,"status":"ok","timestamp":1673625232858,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"uFC3VgqeGhHH","outputId":"6824f1f6-7969-4084-e625-ec8a6ff3a11d"},"outputs":[{"name":"stdout","output_type":"stream","text":["['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","\n","\n","\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n","['NOISE', 'END', 'END', 'END', 'END', 'END']\n"]}],"source":["# Compare the ground truth labels with the predicted labels for a sample of the test set\n","ground_truth = [[tokens_to_syllables[token.item()] for token in prediction] for prediction in test_output_tokens[0:10]]\n","for tag in ground_truth:\n","    print(tag)\n","\n","print(\"\\n\\n\")    \n","predictions = test_predictions(0, 10)\n","for tag in predictions:\n","    print(tag)"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"D5e3YlUd1coi"},"outputs":[],"source":["# Adapted from https://github.com/zszyellow/WER-in-python/blob/master/wer.py\n","def get_phone_error_rate(truth_line, predicted_line):\n","    edit_matrix = np.zeros((len(truth_line) + 1, len(predicted_line) + 1), dtype=np.int8)\n","\n","    for i in range(len(truth_line) + 1):\n","        edit_matrix[i][0] = i\n","    for j in range(len(predicted_line) + 1):\n","        edit_matrix[0][j] = j\n","    for i in range(1, len(truth_line) + 1):\n","        for j in range(1, len(predicted_line) + 1):\n","            if truth_line[i-1] == predicted_line[j-1]:\n","                edit_matrix[i][j] = edit_matrix[i-1][j-1]\n","            else:\n","                substitution_cost = edit_matrix[i-1][j-1] + 1\n","                insertion_cost = edit_matrix[i][j-1] + 1\n","                deletion_cost = edit_matrix[i-1][j] + 1\n","                edit_matrix[i][j] = min(substitution_cost, insertion_cost, deletion_cost)\n","    return edit_matrix[-1, -1] / len(truth_line)"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"zUM80ZMD0h8x"},"outputs":[],"source":["# Copied from https://github.com/zszyellow/WER-in-python/blob/master/wer.py\n","# Explained at https://martin-thoma.com/word-error-rate-calculation/\n","def weighted_phoneme_error_rate(batch_size=128):\n","    total_phone_error_rate = 0\n","    test_batches = math.ceil(len(test_output_tokens) / batch_size)\n","    for batch_idx in range(test_batches):\n","        batch_start = batch_idx * batch_size\n","        batch_end = min(batch_start + batch_size, len(test_output_tokens))\n","\n","        spectrogram_batch = test_spectrograms[batch_start:batch_end]\n","        output_tokens_batch = test_output_tokens[batch_start:batch_end]\n","\n","        _, output = test_batch(encoder, decoder, spectrogram_batch, output_tokens_batch)\n","        predictions = torch.argmax(output, dim=-1).cpu().numpy()\n","\n","        for line_idx in range(len(output_tokens_batch)):\n","            truth_line = output_tokens_batch[line_idx].to(cpu).numpy()\n","            predicted_line = predictions[line_idx]\n","            phoneme_error_rate = get_phone_error_rate(truth_line, predicted_line)\n","            total_phone_error_rate += phoneme_error_rate\n","\n","            print(\"truth line: \", truth_line)\n","            print(\"predicted_line: \", predicted_line)\n","            print(\"PER:\", phoneme_error_rate)\n","            print(\"\\n\")\n","\n","    return total_phone_error_rate / len(test_output_tokens)"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"1wydhmjjJx-F","outputId":"3b1a97b7-3921-4c35-8ca9-e766b6260828"},"outputs":[{"name":"stdout","output_type":"stream","text":["truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 13 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 20  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 20  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 17 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.3076923076923077\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.38461538461538464\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 20  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.46153846153846156\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [25 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.6153846153846154\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 27  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.3076923076923077\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [12 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 20  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 17 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.38461538461538464\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 16 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 17 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 16 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 25  9 13 16 18 30 10  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.3076923076923077\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  9  8 16 18 30 10  0]\n","predicted_line:  [12  4 13 24  2 11  9 17 16 18 30 10  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [26 28  5  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","predicted_line:  [27 20  8 27 27 17  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","predicted_line:  [26 28 19  8 24 11 23 17  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","predicted_line:  [25  7 14  8 24 11 18 30 10  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15 21 20  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 21 20  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 26 28  5 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [25  9 13 16  8 24 11 18 30 10  0  0  0]\n","predicted_line:  [25  9 13 24  8 24 11 18 30 10  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 25  7 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27  4 27  0  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [2 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","predicted_line:  [12  4 13 24  2 29 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","predicted_line:  [26 15 22 21  8 24 11 23 17  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","predicted_line:  [27 27 27 21 10  0  0  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [12  0  0  0  0  0  0  0  0  0  0  0  0]\n","PER: 0.07692307692307693\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","predicted_line:  [12  4 13 24  2 11  6 14 18 30 10  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","predicted_line:  [12  4 13 24  2 15  5 18 30 10  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","predicted_line:  [25 27  8 24 11 27 20  0  0  0  0  0  0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n","truth line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","predicted_line:  [3 0 0 0 0 0 0 0 0 0 0 0 0]\n","PER: 0.0\n","\n","\n"]},{"data":{"text/plain":["0.01778568252556691"]},"execution_count":63,"metadata":{"tags":[]},"output_type":"execute_result"}],"source":["weighted_phoneme_error_rate()"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"elapsed":34,"status":"ok","timestamp":1673625278343,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"R3vScn-EUcdx","outputId":"0f014e98-a75a-4a88-8045-1c9e6508b15c"},"outputs":[{"name":"stdout","output_type":"stream","text":["{1: 2192, 2: 0, 3: 0, 4: 0, 5: 54, 6: 0, 7: 0, 8: 0, 9: 0, 10: 0, 11: 0, 12: 0, 13: 0, 14: 0, 15: 0}\n"]}],"source":["# Print the amount of times token sequences of each length appear in the training set\n","seqs_of_len = {i:0 for i in range(1, 16)}\n","for seq in train_output_tokens.cpu().numpy():\n","    non_trivial_tokens = 0;\n","    for inseq in seq:\n","        if inseq != 0:\n","            non_trivial_tokens+=1\n","        else:\n","            break;\n","\n","    if non_trivial_tokens in seqs_of_len:\n","        seqs_of_len[non_trivial_tokens] += 1\n","    else:\n","        seqs_of_len[non_trivial_tokens] = 1\n","\n","print(seqs_of_len)"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"WYHwkmKO7h8D"},"outputs":[],"source":["# Create a confusion matrix based on the test\n","def create_confusion_matrix(encoder, decoder, batch_size=128):\n","    confusion_matrix = np.zeros([len(syl_to_tokens), len(syl_to_tokens)])\n","\n","    test_batches = math.ceil(len(test_output_tokens) / batch_size)\n","    for batch_idx in range(test_batches):\n","        batch_start = batch_idx * batch_size\n","        batch_end = min(batch_start + batch_size, len(test_output_tokens))\n","\n","        spectrogram_batch = test_spectrograms[batch_start:batch_end]\n","        output_tokens_batch = test_output_tokens[batch_start:batch_end]\n","\n","        _, output = test_batch(encoder, decoder, spectrogram_batch, output_tokens_batch)\n","        predictions = torch.argmax(output, dim=-1).cpu().numpy()\n","\n","        for line_idx in range(len(output_tokens_batch)):\n","            truth_line = output_tokens_batch[line_idx]\n","            predicted_line = predictions[line_idx]\n","\n","            for token_idx in range(len(truth_line)):\n","                true_token = truth_line[token_idx]\n","                predicted_token = predicted_line[token_idx]\n","                confusion_matrix[true_token, predicted_token] += 1\n","\n","    true_label_norm_coefficients = np.reshape(np.repeat(np.sum(confusion_matrix, axis=1), len(confusion_matrix)), (len(confusion_matrix), len(confusion_matrix))) + 1e-12\n","    return confusion_matrix, true_label_norm_coefficients"]},{"cell_type":"code","execution_count":null,"metadata":{"id":"Z05Zp1r_Ig8k"},"outputs":[],"source":["# Plot a confusion matrix\n","def plot_confusion_matrix(confusion_matrix, syllables, figsize=(24,24)):\n","    fig = plt.figure(figsize=figsize)\n","    ax = sns.heatmap(confusion_matrix, annot=True, fmt='.2f', cbar_kws={\"aspect\":50})\n","\n","    ax.set_xticklabels(syllables)\n","    ax.set_yticklabels(syllables)\n","    ax.set_xlabel(\"Predicted syllable\")\n","    ax.set_ylabel(\"True syllable\")\n","    ax.tick_params(\n","        labelbottom=True, labeltop=True, labelleft=True, labelright=True,\n","        bottom=True, top=True, left=True, right=True\n","    )\n","\n","    plt.show()"]},{"cell_type":"code","execution_count":null,"metadata":{"colab":{"base_uri":"https://localhost:8080/","height":1000},"executionInfo":{"elapsed":4972,"status":"ok","timestamp":1673625305109,"user":{"displayName":"Malcolm Ke Win","userId":"08379113201439815447"},"user_tz":-480},"id":"vhCE87Y89Ptw","outputId":"90f1339a-e4f3-4873-9d87-04935b066734"},"outputs":[{"data":{"image/png":"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\n","text/plain":["<Figure size 1728x1728 with 2 Axes>"]},"metadata":{},"output_type":"display_data"}],"source":["confusion_matrix, norm_coefficients = create_confusion_matrix(encoder, decoder)\n","plot_confusion_matrix(confusion_matrix / norm_coefficients, syl_to_tokens.keys())"]}],"metadata":{"accelerator":"GPU","colab":{"machine_shape":"hm","provenance":[{"file_id":"https://github.com/diyism/colab_notebooks/blob/main/Hebrew_AC_voice_activation_(public_version).ipynb","timestamp":1673409384172}]},"kernelspec":{"display_name":"Python 3","name":"python3"}},"nbformat":4,"nbformat_minor":0}