perform_cnn.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {},
      "outputs": [],
      "source": [
        "### Import libraries and packages\n",
        "import os\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import torch\n",
        "from torch.nn import functional \n",
        "from pydub import AudioSegment\n",
        "import librosa"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 86,
      "metadata": {},
      "outputs": [],
      "source": [
        "### Helper function to count the number of files in a directory\n",
        "def count_files(directory):\n",
        "    count = 0\n",
        "    for _, _, files in os.walk(directory):\n",
        "        count += len(files)\n",
        "    return count"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 87,
      "metadata": {},
      "outputs": [],
      "source": [
        "### Class label mappings for 12 different genres\n",
        "MAPPING = {\n",
        "    \"acoustic\" : 0,\n",
        "    \"blues\" : 1,\n",
        "    \"classical\" : 2,\n",
        "    \"country\" : 3,\n",
        "    \"dance\" : 4,\n",
        "    \"edm\" : 5,\n",
        "    \"hip-hop\" : 6,\n",
        "    \"indie\" : 7,\n",
        "    \"jazz\" : 8,\n",
        "    \"pop\" : 9,\n",
        "    \"rock\" : 10,\n",
        "    \"world-music\" : 11\n",
        "}\n",
        "\n",
        "IMG_FOLDER = \"spectrogram\"\n",
        "num_files = count_files(IMG_FOLDER)\n",
        "img_height = 65\n",
        "img_width = 86\n",
        "genres = os.listdir(IMG_FOLDER)\n",
        "\n",
        "### Initialize empty images and labels array \n",
        "images = np.zeros(shape = (num_files, 4, img_height, img_width))\n",
        "labels = np.zeros(shape = num_files)\n",
        "\n",
        "### Fill images array with melspectrograms and labels array with their respective genre\n",
        "idx = 0\n",
        "for root, directories, contents in os.walk(IMG_FOLDER):\n",
        "    if root == IMG_FOLDER: continue\n",
        "    genre = root.split('/')[1]\n",
        "    if genre in MAPPING:\n",
        "        for img_path in contents:\n",
        "            image = plt.imread(root + \"/\" + img_path)\n",
        "            \n",
        "            ## Convert to tensor and check the shape\n",
        "            spec = torch.from_numpy(image)\n",
        "\n",
        "            ## Move third dimension (color channels) to the first dimension\n",
        "            spec2 = torch.movedim(spec, source=2, destination=0)\n",
        "\n",
        "            ## Add an empty first dimension, putting our tensor into the standard format\n",
        "            spec_final = torch.unsqueeze(spec2, dim=0)\n",
        "\n",
        "            spec_resized = functional.interpolate(spec_final, size = (img_height,img_width))\n",
        "            \n",
        "            images[idx] = spec_resized\n",
        "            labels[idx] = MAPPING[genre]\n",
        "            idx += 1\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 88,
      "metadata": {},
      "outputs": [],
      "source": [
        "## Train-test split\n",
        "from sklearn.model_selection import train_test_split\n",
        "train_X, test_X, train_y, test_y = train_test_split(images, labels, test_size=0.25, random_state=5)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 148,
      "metadata": {},
      "outputs": [],
      "source": [
        "from torch import nn\n",
        "\n",
        "class my_net(nn.Module):\n",
        "    \n",
        "    ## Constructor commands\n",
        "    def __init__(self):\n",
        "        super(my_net, self).__init__()\n",
        "        \n",
        "        ## Define architecture\n",
        "        self.conv_stack = nn.Sequential(\n",
        "            nn.Conv2d(4,8,3,1),\n",
        "            nn.ReLU(),\n",
        "            nn.MaxPool2d(2, 2),\n",
        "            nn.Conv2d(8,16,3,1),\n",
        "            nn.ReLU(),\n",
        "            nn.MaxPool2d(2, 2),\n",
        "            nn.Flatten(),\n",
        "            nn.Linear(4480, 200),\n",
        "            nn.ReLU(),\n",
        "            nn.Linear(200, 12)\n",
        "        )\n",
        "    \n",
        "    ## Function to generate predictions\n",
        "    def forward(self, x):\n",
        "        scores = self.conv_stack(x)\n",
        "        return scores"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 149,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "torch.Size([732, 4, 65, 89])\n"
          ]
        }
      ],
      "source": [
        "## Hyperparms\n",
        "epochs = 1000\n",
        "lrate = 0.01\n",
        "bsize = 10\n",
        "\n",
        "## Cost Function\n",
        "cost_fn = nn.CrossEntropyLoss()\n",
        "\n",
        "## Intialize the model\n",
        "net = my_net()\n",
        "\n",
        "## Optimizer (Stochastic Gradient Descent)\n",
        "optimizer = torch.optim.SGD(net.parameters(), lr=lrate)\n",
        "\n",
        "## Make DataLoader\n",
        "from torch.utils.data import DataLoader, TensorDataset\n",
        "y_tensor = torch.Tensor(train_y)\n",
        "X_tensor = torch.from_numpy(train_X)\n",
        "print(X_tensor.shape)\n",
        "train_loader = DataLoader(\n",
        "  TensorDataset(\n",
        "    X_tensor.type(torch.FloatTensor), \n",
        "    y_tensor.type(torch.LongTensor)),\n",
        "  batch_size=bsize\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 150,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch: 0 Cost: 183.85356402397156\n",
            "Epoch: 1 Cost: 183.8362078666687\n",
            "Epoch: 2 Cost: 183.8191523551941\n",
            "Epoch: 3 Cost: 183.80222034454346\n",
            "Epoch: 4 Cost: 183.78484582901\n",
            "Epoch: 5 Cost: 183.76610374450684\n",
            "Epoch: 6 Cost: 183.74496221542358\n",
            "Epoch: 7 Cost: 183.72105503082275\n",
            "Epoch: 8 Cost: 183.6935839653015\n",
            "Epoch: 9 Cost: 183.6618218421936\n",
            "Epoch: 10 Cost: 183.62481427192688\n",
            "Epoch: 11 Cost: 183.58113288879395\n",
            "Epoch: 12 Cost: 183.52945256233215\n",
            "Epoch: 13 Cost: 183.46828055381775\n",
            "Epoch: 14 Cost: 183.39730763435364\n",
            "Epoch: 15 Cost: 183.3181073665619\n",
            "Epoch: 16 Cost: 183.23629546165466\n",
            "Epoch: 17 Cost: 183.16125059127808\n",
            "Epoch: 18 Cost: 183.10108590126038\n",
            "Epoch: 19 Cost: 183.0552840232849\n",
            "Epoch: 20 Cost: 183.01918601989746\n",
            "Epoch: 21 Cost: 182.9879732131958\n",
            "Epoch: 22 Cost: 182.9617416858673\n",
            "Epoch: 23 Cost: 182.94257950782776\n",
            "Epoch: 24 Cost: 182.92515540122986\n",
            "Epoch: 25 Cost: 182.91167092323303\n",
            "Epoch: 26 Cost: 182.90095043182373\n",
            "Epoch: 27 Cost: 182.89160585403442\n",
            "Epoch: 28 Cost: 182.88334369659424\n",
            "Epoch: 29 Cost: 182.87457132339478\n",
            "Epoch: 30 Cost: 182.8645739555359\n",
            "Epoch: 31 Cost: 182.8541567325592\n",
            "Epoch: 32 Cost: 182.84156107902527\n",
            "Epoch: 33 Cost: 182.82705283164978\n",
            "Epoch: 34 Cost: 182.81079196929932\n",
            "Epoch: 35 Cost: 182.79178524017334\n",
            "Epoch: 36 Cost: 182.76517009735107\n",
            "Epoch: 37 Cost: 182.73338627815247\n",
            "Epoch: 38 Cost: 182.69077610969543\n",
            "Epoch: 39 Cost: 182.63495302200317\n",
            "Epoch: 40 Cost: 182.55757355690002\n",
            "Epoch: 41 Cost: 182.44199585914612\n",
            "Epoch: 42 Cost: 182.26867389678955\n",
            "Epoch: 43 Cost: 182.01365447044373\n",
            "Epoch: 44 Cost: 181.69485330581665\n",
            "Epoch: 45 Cost: 181.3477942943573\n",
            "Epoch: 46 Cost: 181.03264927864075\n",
            "Epoch: 47 Cost: 180.76956391334534\n",
            "Epoch: 48 Cost: 180.5626084804535\n",
            "Epoch: 49 Cost: 180.3883922100067\n",
            "Epoch: 50 Cost: 180.23811435699463\n",
            "Epoch: 51 Cost: 180.10600233078003\n",
            "Epoch: 52 Cost: 179.9847650527954\n",
            "Epoch: 53 Cost: 179.867901802063\n",
            "Epoch: 54 Cost: 179.75812911987305\n",
            "Epoch: 55 Cost: 179.6584119796753\n",
            "Epoch: 56 Cost: 179.5597381591797\n",
            "Epoch: 57 Cost: 179.46918749809265\n",
            "Epoch: 58 Cost: 179.37413454055786\n",
            "Epoch: 59 Cost: 179.28909158706665\n",
            "Epoch: 60 Cost: 179.21096992492676\n",
            "Epoch: 61 Cost: 179.12483477592468\n",
            "Epoch: 62 Cost: 179.03977179527283\n",
            "Epoch: 63 Cost: 178.96081399917603\n",
            "Epoch: 64 Cost: 178.8757245540619\n",
            "Epoch: 65 Cost: 178.79636883735657\n",
            "Epoch: 66 Cost: 178.71273827552795\n",
            "Epoch: 67 Cost: 178.6273033618927\n",
            "Epoch: 68 Cost: 178.5400938987732\n",
            "Epoch: 69 Cost: 178.45868396759033\n",
            "Epoch: 70 Cost: 178.36972188949585\n",
            "Epoch: 71 Cost: 178.28119945526123\n",
            "Epoch: 72 Cost: 178.19344830513\n",
            "Epoch: 73 Cost: 178.11058139801025\n",
            "Epoch: 74 Cost: 178.02611804008484\n",
            "Epoch: 75 Cost: 177.94105672836304\n",
            "Epoch: 76 Cost: 177.85900783538818\n",
            "Epoch: 77 Cost: 177.77367448806763\n",
            "Epoch: 78 Cost: 177.6859164237976\n",
            "Epoch: 79 Cost: 177.60468006134033\n",
            "Epoch: 80 Cost: 177.52244472503662\n",
            "Epoch: 81 Cost: 177.44391226768494\n",
            "Epoch: 82 Cost: 177.36912751197815\n",
            "Epoch: 83 Cost: 177.2862560749054\n",
            "Epoch: 84 Cost: 177.19805717468262\n",
            "Epoch: 85 Cost: 177.11705422401428\n",
            "Epoch: 86 Cost: 177.03826665878296\n",
            "Epoch: 87 Cost: 176.9743115901947\n",
            "Epoch: 88 Cost: 176.89895796775818\n",
            "Epoch: 89 Cost: 176.83534288406372\n",
            "Epoch: 90 Cost: 176.76038074493408\n",
            "Epoch: 91 Cost: 176.70073986053467\n",
            "Epoch: 92 Cost: 176.64839220046997\n",
            "Epoch: 93 Cost: 176.58722710609436\n",
            "Epoch: 94 Cost: 176.5277829170227\n",
            "Epoch: 95 Cost: 176.4685299396515\n",
            "Epoch: 96 Cost: 176.39306831359863\n",
            "Epoch: 97 Cost: 176.34501361846924\n",
            "Epoch: 98 Cost: 176.2939567565918\n",
            "Epoch: 99 Cost: 176.23361778259277\n",
            "Epoch: 100 Cost: 176.17549753189087\n",
            "Epoch: 101 Cost: 176.12699449062347\n",
            "Epoch: 102 Cost: 176.07486629486084\n",
            "Epoch: 103 Cost: 176.04294097423553\n",
            "Epoch: 104 Cost: 176.0172791481018\n",
            "Epoch: 105 Cost: 175.96140897274017\n",
            "Epoch: 106 Cost: 175.91358411312103\n",
            "Epoch: 107 Cost: 175.87081718444824\n",
            "Epoch: 108 Cost: 175.81265330314636\n",
            "Epoch: 109 Cost: 175.76379823684692\n",
            "Epoch: 110 Cost: 175.71423053741455\n",
            "Epoch: 111 Cost: 175.64352333545685\n",
            "Epoch: 112 Cost: 175.58379471302032\n",
            "Epoch: 113 Cost: 175.53856694698334\n",
            "Epoch: 114 Cost: 175.4911905527115\n",
            "Epoch: 115 Cost: 175.43287229537964\n",
            "Epoch: 116 Cost: 175.36592268943787\n",
            "Epoch: 117 Cost: 175.31036067008972\n",
            "Epoch: 118 Cost: 175.26552093029022\n",
            "Epoch: 119 Cost: 175.21161270141602\n",
            "Epoch: 120 Cost: 175.17339718341827\n",
            "Epoch: 121 Cost: 175.121928691864\n",
            "Epoch: 122 Cost: 175.06649041175842\n",
            "Epoch: 123 Cost: 175.04061841964722\n",
            "Epoch: 124 Cost: 174.9812754392624\n",
            "Epoch: 125 Cost: 174.9385108947754\n",
            "Epoch: 126 Cost: 174.83936381340027\n",
            "Epoch: 127 Cost: 174.7568416595459\n",
            "Epoch: 128 Cost: 174.70303463935852\n",
            "Epoch: 129 Cost: 174.66573643684387\n",
            "Epoch: 130 Cost: 174.63542199134827\n",
            "Epoch: 131 Cost: 174.52990794181824\n",
            "Epoch: 132 Cost: 174.5097758769989\n",
            "Epoch: 133 Cost: 174.47547578811646\n",
            "Epoch: 134 Cost: 174.44408917427063\n",
            "Epoch: 135 Cost: 174.36419904232025\n",
            "Epoch: 136 Cost: 174.30218029022217\n",
            "Epoch: 137 Cost: 174.29006242752075\n",
            "Epoch: 138 Cost: 174.27079558372498\n",
            "Epoch: 139 Cost: 174.17687487602234\n",
            "Epoch: 140 Cost: 174.3170541524887\n",
            "Epoch: 141 Cost: 174.3458435535431\n",
            "Epoch: 142 Cost: 174.15008199214935\n",
            "Epoch: 143 Cost: 174.31180143356323\n",
            "Epoch: 144 Cost: 174.0283966064453\n",
            "Epoch: 145 Cost: 174.33533799648285\n",
            "Epoch: 146 Cost: 174.23943281173706\n",
            "Epoch: 147 Cost: 174.3754916191101\n",
            "Epoch: 148 Cost: 173.93647468090057\n",
            "Epoch: 149 Cost: 173.82683062553406\n",
            "Epoch: 150 Cost: 173.88112092018127\n",
            "Epoch: 151 Cost: 173.88781464099884\n",
            "Epoch: 152 Cost: 173.1414588689804\n",
            "Epoch: 153 Cost: 173.23481690883636\n",
            "Epoch: 154 Cost: 173.22379553318024\n",
            "Epoch: 155 Cost: 173.16132378578186\n",
            "Epoch: 156 Cost: 173.41075384616852\n",
            "Epoch: 157 Cost: 173.70143055915833\n",
            "Epoch: 158 Cost: 172.9539259672165\n",
            "Epoch: 159 Cost: 173.10809552669525\n",
            "Epoch: 160 Cost: 172.9661487340927\n",
            "Epoch: 161 Cost: 174.1069779396057\n",
            "Epoch: 162 Cost: 172.68308699131012\n",
            "Epoch: 163 Cost: 172.7025501728058\n",
            "Epoch: 164 Cost: 172.89189755916595\n",
            "Epoch: 165 Cost: 172.85845685005188\n",
            "Epoch: 166 Cost: 172.73936009407043\n",
            "Epoch: 167 Cost: 172.69096386432648\n",
            "Epoch: 168 Cost: 172.49171936511993\n",
            "Epoch: 169 Cost: 172.5300680398941\n",
            "Epoch: 170 Cost: 172.89259827136993\n",
            "Epoch: 171 Cost: 172.55922961235046\n",
            "Epoch: 172 Cost: 173.2172416448593\n",
            "Epoch: 173 Cost: 174.5198459625244\n",
            "Epoch: 174 Cost: 173.30239617824554\n",
            "Epoch: 175 Cost: 172.69810330867767\n",
            "Epoch: 176 Cost: 172.9280549287796\n",
            "Epoch: 177 Cost: 172.27440750598907\n",
            "Epoch: 178 Cost: 172.41123032569885\n",
            "Epoch: 179 Cost: 171.95485055446625\n",
            "Epoch: 180 Cost: 172.70177245140076\n",
            "Epoch: 181 Cost: 172.7033212184906\n",
            "Epoch: 182 Cost: 171.69135200977325\n",
            "Epoch: 183 Cost: 171.7517009973526\n",
            "Epoch: 184 Cost: 172.7609385251999\n",
            "Epoch: 185 Cost: 172.68102979660034\n",
            "Epoch: 186 Cost: 172.34939861297607\n",
            "Epoch: 187 Cost: 172.09152054786682\n",
            "Epoch: 188 Cost: 172.17836499214172\n",
            "Epoch: 189 Cost: 172.27420663833618\n",
            "Epoch: 190 Cost: 172.03208303451538\n",
            "Epoch: 191 Cost: 171.7543145418167\n",
            "Epoch: 192 Cost: 172.50581777095795\n",
            "Epoch: 193 Cost: 172.80765116214752\n",
            "Epoch: 194 Cost: 171.73993813991547\n",
            "Epoch: 195 Cost: 171.74918282032013\n",
            "Epoch: 196 Cost: 172.30448043346405\n",
            "Epoch: 197 Cost: 172.04997038841248\n",
            "Epoch: 198 Cost: 171.51629519462585\n",
            "Epoch: 199 Cost: 171.81979262828827\n",
            "Epoch: 200 Cost: 171.2822494506836\n",
            "Epoch: 201 Cost: 172.34689366817474\n",
            "Epoch: 202 Cost: 171.4655898809433\n",
            "Epoch: 203 Cost: 171.90004670619965\n",
            "Epoch: 204 Cost: 171.3497543334961\n",
            "Epoch: 205 Cost: 171.83105444908142\n",
            "Epoch: 206 Cost: 171.04766499996185\n",
            "Epoch: 207 Cost: 171.95363187789917\n",
            "Epoch: 208 Cost: 171.25365698337555\n",
            "Epoch: 209 Cost: 172.17224299907684\n",
            "Epoch: 210 Cost: 171.77523040771484\n",
            "Epoch: 211 Cost: 171.74038064479828\n",
            "Epoch: 212 Cost: 172.35999608039856\n",
            "Epoch: 213 Cost: 170.998366355896\n",
            "Epoch: 214 Cost: 171.24394404888153\n",
            "Epoch: 215 Cost: 171.1616849899292\n",
            "Epoch: 216 Cost: 172.0586338043213\n",
            "Epoch: 217 Cost: 171.24454987049103\n",
            "Epoch: 218 Cost: 171.15907514095306\n",
            "Epoch: 219 Cost: 170.64978420734406\n",
            "Epoch: 220 Cost: 171.23112654685974\n",
            "Epoch: 221 Cost: 170.47526741027832\n",
            "Epoch: 222 Cost: 170.23785614967346\n",
            "Epoch: 223 Cost: 171.6018786430359\n",
            "Epoch: 224 Cost: 171.41070020198822\n",
            "Epoch: 225 Cost: 171.2275391817093\n",
            "Epoch: 226 Cost: 171.00815641880035\n",
            "Epoch: 227 Cost: 172.49639666080475\n",
            "Epoch: 228 Cost: 170.71232163906097\n",
            "Epoch: 229 Cost: 172.14970135688782\n",
            "Epoch: 230 Cost: 171.08999860286713\n",
            "Epoch: 231 Cost: 172.61816716194153\n",
            "Epoch: 232 Cost: 172.64741563796997\n",
            "Epoch: 233 Cost: 171.1637727022171\n",
            "Epoch: 234 Cost: 172.39744234085083\n",
            "Epoch: 235 Cost: 172.89809489250183\n",
            "Epoch: 236 Cost: 172.15848088264465\n",
            "Epoch: 237 Cost: 170.4900028705597\n",
            "Epoch: 238 Cost: 170.67278110980988\n",
            "Epoch: 239 Cost: 174.375342130661\n",
            "Epoch: 240 Cost: 170.49918246269226\n",
            "Epoch: 241 Cost: 170.9732631444931\n",
            "Epoch: 242 Cost: 171.83836460113525\n",
            "Epoch: 243 Cost: 172.7183063030243\n",
            "Epoch: 244 Cost: 170.02453112602234\n",
            "Epoch: 245 Cost: 170.1325067281723\n",
            "Epoch: 246 Cost: 170.30352318286896\n",
            "Epoch: 247 Cost: 170.1222584247589\n",
            "Epoch: 248 Cost: 171.28423380851746\n",
            "Epoch: 249 Cost: 170.80360746383667\n",
            "Epoch: 250 Cost: 170.4694048166275\n",
            "Epoch: 251 Cost: 170.35825181007385\n",
            "Epoch: 252 Cost: 170.2880116701126\n",
            "Epoch: 253 Cost: 169.68296539783478\n",
            "Epoch: 254 Cost: 170.26650381088257\n",
            "Epoch: 255 Cost: 169.77079093456268\n",
            "Epoch: 256 Cost: 170.0774462223053\n",
            "Epoch: 257 Cost: 169.5598340034485\n",
            "Epoch: 258 Cost: 169.72836136817932\n",
            "Epoch: 259 Cost: 169.37635266780853\n",
            "Epoch: 260 Cost: 169.002676486969\n",
            "Epoch: 261 Cost: 169.82575833797455\n",
            "Epoch: 262 Cost: 170.14177453517914\n",
            "Epoch: 263 Cost: 169.60202586650848\n",
            "Epoch: 264 Cost: 170.18095231056213\n",
            "Epoch: 265 Cost: 171.11076200008392\n",
            "Epoch: 266 Cost: 169.48151290416718\n",
            "Epoch: 267 Cost: 169.84305238723755\n",
            "Epoch: 268 Cost: 173.5270230770111\n",
            "Epoch: 269 Cost: 172.86148643493652\n",
            "Epoch: 270 Cost: 169.70174098014832\n",
            "Epoch: 271 Cost: 169.49678468704224\n",
            "Epoch: 272 Cost: 171.48575103282928\n",
            "Epoch: 273 Cost: 169.40760946273804\n",
            "Epoch: 274 Cost: 169.44288730621338\n",
            "Epoch: 275 Cost: 168.969300031662\n",
            "Epoch: 276 Cost: 169.34677004814148\n",
            "Epoch: 277 Cost: 169.25395214557648\n",
            "Epoch: 278 Cost: 169.01714837551117\n",
            "Epoch: 279 Cost: 169.007120013237\n",
            "Epoch: 280 Cost: 168.64750349521637\n",
            "Epoch: 281 Cost: 168.85011279582977\n",
            "Epoch: 282 Cost: 169.10370290279388\n",
            "Epoch: 283 Cost: 168.75911903381348\n",
            "Epoch: 284 Cost: 170.90424978733063\n",
            "Epoch: 285 Cost: 169.1680645942688\n",
            "Epoch: 286 Cost: 168.82224106788635\n",
            "Epoch: 287 Cost: 170.47051191329956\n",
            "Epoch: 288 Cost: 168.63740181922913\n",
            "Epoch: 289 Cost: 168.7063933610916\n",
            "Epoch: 290 Cost: 169.30052852630615\n",
            "Epoch: 291 Cost: 168.77990782260895\n",
            "Epoch: 292 Cost: 170.57866024971008\n",
            "Epoch: 293 Cost: 168.8115358352661\n",
            "Epoch: 294 Cost: 168.8919974565506\n",
            "Epoch: 295 Cost: 168.8910870552063\n",
            "Epoch: 296 Cost: 168.7512445449829\n",
            "Epoch: 297 Cost: 170.4144276380539\n",
            "Epoch: 298 Cost: 170.94629275798798\n",
            "Epoch: 299 Cost: 169.8474212884903\n",
            "Epoch: 300 Cost: 168.90271389484406\n",
            "Epoch: 301 Cost: 169.75821816921234\n",
            "Epoch: 302 Cost: 171.34018898010254\n",
            "Epoch: 303 Cost: 168.67591261863708\n",
            "Epoch: 304 Cost: 168.45636224746704\n",
            "Epoch: 305 Cost: 169.61331582069397\n",
            "Epoch: 306 Cost: 168.33449125289917\n",
            "Epoch: 307 Cost: 168.3752818107605\n",
            "Epoch: 308 Cost: 169.72411346435547\n",
            "Epoch: 309 Cost: 168.52574336528778\n",
            "Epoch: 310 Cost: 168.02231979370117\n",
            "Epoch: 311 Cost: 170.99043703079224\n",
            "Epoch: 312 Cost: 168.6332575082779\n",
            "Epoch: 313 Cost: 168.52600145339966\n",
            "Epoch: 314 Cost: 170.30810225009918\n",
            "Epoch: 315 Cost: 168.11844050884247\n",
            "Epoch: 316 Cost: 169.16809558868408\n",
            "Epoch: 317 Cost: 168.46483314037323\n",
            "Epoch: 318 Cost: 168.81830477714539\n",
            "Epoch: 319 Cost: 168.1637749671936\n",
            "Epoch: 320 Cost: 170.3810477256775\n",
            "Epoch: 321 Cost: 168.94888353347778\n",
            "Epoch: 322 Cost: 167.99547064304352\n",
            "Epoch: 323 Cost: 167.8568059206009\n",
            "Epoch: 324 Cost: 168.04417884349823\n",
            "Epoch: 325 Cost: 168.83841598033905\n",
            "Epoch: 326 Cost: 168.40031909942627\n",
            "Epoch: 327 Cost: 168.05615317821503\n",
            "Epoch: 328 Cost: 167.99546945095062\n",
            "Epoch: 329 Cost: 169.24959540367126\n",
            "Epoch: 330 Cost: 169.33906960487366\n",
            "Epoch: 331 Cost: 168.1813395023346\n",
            "Epoch: 332 Cost: 168.14882493019104\n",
            "Epoch: 333 Cost: 167.9406180381775\n",
            "Epoch: 334 Cost: 167.62798261642456\n",
            "Epoch: 335 Cost: 168.20046496391296\n",
            "Epoch: 336 Cost: 168.418390750885\n",
            "Epoch: 337 Cost: 167.1577467918396\n",
            "Epoch: 338 Cost: 169.11635422706604\n",
            "Epoch: 339 Cost: 167.93345606327057\n",
            "Epoch: 340 Cost: 167.56764209270477\n",
            "Epoch: 341 Cost: 167.56038212776184\n",
            "Epoch: 342 Cost: 167.93041133880615\n",
            "Epoch: 343 Cost: 168.39125418663025\n",
            "Epoch: 344 Cost: 167.40701627731323\n",
            "Epoch: 345 Cost: 170.13030111789703\n",
            "Epoch: 346 Cost: 167.65610098838806\n",
            "Epoch: 347 Cost: 167.7101687192917\n",
            "Epoch: 348 Cost: 168.3802958726883\n",
            "Epoch: 349 Cost: 167.17068123817444\n",
            "Epoch: 350 Cost: 166.730539560318\n",
            "Epoch: 351 Cost: 167.56158471107483\n",
            "Epoch: 352 Cost: 167.22560858726501\n",
            "Epoch: 353 Cost: 166.72334361076355\n",
            "Epoch: 354 Cost: 167.70928156375885\n",
            "Epoch: 355 Cost: 168.20393574237823\n",
            "Epoch: 356 Cost: 166.51992404460907\n",
            "Epoch: 357 Cost: 166.58164620399475\n",
            "Epoch: 358 Cost: 166.39627885818481\n",
            "Epoch: 359 Cost: 166.22719538211823\n",
            "Epoch: 360 Cost: 167.3994209766388\n",
            "Epoch: 361 Cost: 166.49329614639282\n",
            "Epoch: 362 Cost: 167.96435749530792\n",
            "Epoch: 363 Cost: 166.817391872406\n",
            "Epoch: 364 Cost: 166.4517457485199\n",
            "Epoch: 365 Cost: 166.91463196277618\n",
            "Epoch: 366 Cost: 166.9126180410385\n",
            "Epoch: 367 Cost: 167.28195464611053\n",
            "Epoch: 368 Cost: 167.00870382785797\n",
            "Epoch: 369 Cost: 166.92318034172058\n",
            "Epoch: 370 Cost: 166.24723482131958\n",
            "Epoch: 371 Cost: 166.4641773700714\n",
            "Epoch: 372 Cost: 166.80622279644012\n",
            "Epoch: 373 Cost: 167.39119935035706\n",
            "Epoch: 374 Cost: 166.91728949546814\n",
            "Epoch: 375 Cost: 166.94484984874725\n",
            "Epoch: 376 Cost: 166.5013632774353\n",
            "Epoch: 377 Cost: 166.5440717935562\n",
            "Epoch: 378 Cost: 166.40037941932678\n",
            "Epoch: 379 Cost: 166.58884167671204\n",
            "Epoch: 380 Cost: 166.72828805446625\n",
            "Epoch: 381 Cost: 166.52221822738647\n",
            "Epoch: 382 Cost: 166.79693031311035\n",
            "Epoch: 383 Cost: 166.38118505477905\n",
            "Epoch: 384 Cost: 166.55474185943604\n",
            "Epoch: 385 Cost: 166.1014746427536\n",
            "Epoch: 386 Cost: 167.30153262615204\n",
            "Epoch: 387 Cost: 166.8818712234497\n",
            "Epoch: 388 Cost: 166.4350814819336\n",
            "Epoch: 389 Cost: 166.19020557403564\n",
            "Epoch: 390 Cost: 166.05459260940552\n",
            "Epoch: 391 Cost: 167.20738577842712\n",
            "Epoch: 392 Cost: 166.49651980400085\n",
            "Epoch: 393 Cost: 166.39886736869812\n",
            "Epoch: 394 Cost: 166.0132611989975\n",
            "Epoch: 395 Cost: 165.82430851459503\n",
            "Epoch: 396 Cost: 165.8358850479126\n",
            "Epoch: 397 Cost: 165.968057513237\n",
            "Epoch: 398 Cost: 166.08167898654938\n",
            "Epoch: 399 Cost: 165.70008397102356\n",
            "Epoch: 400 Cost: 165.92255795001984\n",
            "Epoch: 401 Cost: 166.25305593013763\n",
            "Epoch: 402 Cost: 165.40910828113556\n",
            "Epoch: 403 Cost: 165.2733210325241\n",
            "Epoch: 404 Cost: 165.6226304769516\n",
            "Epoch: 405 Cost: 165.72351694107056\n",
            "Epoch: 406 Cost: 165.5335178375244\n",
            "Epoch: 407 Cost: 165.4010888338089\n",
            "Epoch: 408 Cost: 165.26482796669006\n",
            "Epoch: 409 Cost: 165.14354598522186\n",
            "Epoch: 410 Cost: 165.6733272075653\n",
            "Epoch: 411 Cost: 165.55184698104858\n",
            "Epoch: 412 Cost: 165.9768602848053\n",
            "Epoch: 413 Cost: 166.010596036911\n",
            "Epoch: 414 Cost: 165.47520792484283\n",
            "Epoch: 415 Cost: 165.5591241121292\n",
            "Epoch: 416 Cost: 165.2720823287964\n",
            "Epoch: 417 Cost: 165.32866990566254\n",
            "Epoch: 418 Cost: 165.1625655889511\n",
            "Epoch: 419 Cost: 165.20479500293732\n",
            "Epoch: 420 Cost: 165.12748098373413\n",
            "Epoch: 421 Cost: 165.08213937282562\n",
            "Epoch: 422 Cost: 165.33473825454712\n",
            "Epoch: 423 Cost: 165.2628092765808\n",
            "Epoch: 424 Cost: 165.21375703811646\n",
            "Epoch: 425 Cost: 165.08893811702728\n",
            "Epoch: 426 Cost: 164.92620372772217\n",
            "Epoch: 427 Cost: 165.00080752372742\n",
            "Epoch: 428 Cost: 164.90490007400513\n",
            "Epoch: 429 Cost: 165.3855255842209\n",
            "Epoch: 430 Cost: 165.55733108520508\n",
            "Epoch: 431 Cost: 165.1970490217209\n",
            "Epoch: 432 Cost: 165.14616787433624\n",
            "Epoch: 433 Cost: 165.8485186100006\n",
            "Epoch: 434 Cost: 165.0695321559906\n",
            "Epoch: 435 Cost: 165.0490701198578\n",
            "Epoch: 436 Cost: 165.39702594280243\n",
            "Epoch: 437 Cost: 165.06248319149017\n",
            "Epoch: 438 Cost: 165.08018672466278\n",
            "Epoch: 439 Cost: 164.90848398208618\n",
            "Epoch: 440 Cost: 164.93800032138824\n",
            "Epoch: 441 Cost: 164.96561217308044\n",
            "Epoch: 442 Cost: 165.1011769771576\n",
            "Epoch: 443 Cost: 165.20752120018005\n",
            "Epoch: 444 Cost: 164.9397188425064\n",
            "Epoch: 445 Cost: 164.96097922325134\n",
            "Epoch: 446 Cost: 164.7395441532135\n",
            "Epoch: 447 Cost: 164.74518465995789\n",
            "Epoch: 448 Cost: 164.82806730270386\n",
            "Epoch: 449 Cost: 164.56762385368347\n",
            "Epoch: 450 Cost: 164.5817186832428\n",
            "Epoch: 451 Cost: 164.409450173378\n",
            "Epoch: 452 Cost: 165.12293875217438\n",
            "Epoch: 453 Cost: 164.5908088684082\n",
            "Epoch: 454 Cost: 164.46390092372894\n",
            "Epoch: 455 Cost: 164.6990122795105\n",
            "Epoch: 456 Cost: 164.8844907283783\n",
            "Epoch: 457 Cost: 164.93334650993347\n",
            "Epoch: 458 Cost: 164.5958367586136\n",
            "Epoch: 459 Cost: 164.5641416311264\n",
            "Epoch: 460 Cost: 164.3162237405777\n",
            "Epoch: 461 Cost: 165.38378250598907\n",
            "Epoch: 462 Cost: 165.08474683761597\n",
            "Epoch: 463 Cost: 164.6777812242508\n",
            "Epoch: 464 Cost: 164.43961465358734\n",
            "Epoch: 465 Cost: 164.363503575325\n",
            "Epoch: 466 Cost: 164.46637523174286\n",
            "Epoch: 467 Cost: 164.26871526241302\n",
            "Epoch: 468 Cost: 165.0571769475937\n",
            "Epoch: 469 Cost: 164.4701566696167\n",
            "Epoch: 470 Cost: 164.4354864358902\n",
            "Epoch: 471 Cost: 164.26774048805237\n",
            "Epoch: 472 Cost: 164.36366307735443\n",
            "Epoch: 473 Cost: 164.34149181842804\n",
            "Epoch: 474 Cost: 164.587256193161\n",
            "Epoch: 475 Cost: 164.17584204673767\n",
            "Epoch: 476 Cost: 164.7240446805954\n",
            "Epoch: 477 Cost: 164.19241166114807\n",
            "Epoch: 478 Cost: 164.50202560424805\n",
            "Epoch: 479 Cost: 164.2221484184265\n",
            "Epoch: 480 Cost: 164.66131246089935\n",
            "Epoch: 481 Cost: 163.89270567893982\n",
            "Epoch: 482 Cost: 164.25711512565613\n",
            "Epoch: 483 Cost: 164.005065202713\n",
            "Epoch: 484 Cost: 164.2677354812622\n",
            "Epoch: 485 Cost: 164.1715099811554\n",
            "Epoch: 486 Cost: 164.48139119148254\n",
            "Epoch: 487 Cost: 164.355628490448\n",
            "Epoch: 488 Cost: 163.95659244060516\n",
            "Epoch: 489 Cost: 163.88628661632538\n",
            "Epoch: 490 Cost: 163.9539793729782\n",
            "Epoch: 491 Cost: 163.9513862133026\n",
            "Epoch: 492 Cost: 163.86447298526764\n",
            "Epoch: 493 Cost: 164.074361205101\n",
            "Epoch: 494 Cost: 163.76894116401672\n",
            "Epoch: 495 Cost: 164.3930492401123\n",
            "Epoch: 496 Cost: 164.452742934227\n",
            "Epoch: 497 Cost: 164.01832616329193\n",
            "Epoch: 498 Cost: 163.98625874519348\n",
            "Epoch: 499 Cost: 163.91141295433044\n",
            "Epoch: 500 Cost: 163.98305022716522\n",
            "Epoch: 501 Cost: 163.89723241329193\n",
            "Epoch: 502 Cost: 163.88030314445496\n",
            "Epoch: 503 Cost: 163.95176351070404\n",
            "Epoch: 504 Cost: 163.8585867881775\n",
            "Epoch: 505 Cost: 163.84129321575165\n",
            "Epoch: 506 Cost: 163.84975957870483\n",
            "Epoch: 507 Cost: 163.9488959312439\n",
            "Epoch: 508 Cost: 163.8384119272232\n",
            "Epoch: 509 Cost: 163.83513069152832\n",
            "Epoch: 510 Cost: 163.8367154598236\n",
            "Epoch: 511 Cost: 163.93342673778534\n",
            "Epoch: 512 Cost: 163.92403900623322\n",
            "Epoch: 513 Cost: 163.87118136882782\n",
            "Epoch: 514 Cost: 163.97593331336975\n",
            "Epoch: 515 Cost: 164.057342171669\n",
            "Epoch: 516 Cost: 164.01008546352386\n",
            "Epoch: 517 Cost: 163.82606160640717\n",
            "Epoch: 518 Cost: 163.91702342033386\n",
            "Epoch: 519 Cost: 163.80003130435944\n",
            "Epoch: 520 Cost: 163.89434564113617\n",
            "Epoch: 521 Cost: 163.76756525039673\n",
            "Epoch: 522 Cost: 163.73245859146118\n",
            "Epoch: 523 Cost: 163.67763721942902\n",
            "Epoch: 524 Cost: 163.77168130874634\n",
            "Epoch: 525 Cost: 163.79625701904297\n",
            "Epoch: 526 Cost: 163.86795604228973\n",
            "Epoch: 527 Cost: 163.82156419754028\n",
            "Epoch: 528 Cost: 163.77539777755737\n",
            "Epoch: 529 Cost: 163.79805028438568\n",
            "Epoch: 530 Cost: 164.26143372058868\n",
            "Epoch: 531 Cost: 164.06338763237\n",
            "Epoch: 532 Cost: 164.00708532333374\n",
            "Epoch: 533 Cost: 163.72579908370972\n",
            "Epoch: 534 Cost: 163.96412706375122\n",
            "Epoch: 535 Cost: 163.7998435497284\n",
            "Epoch: 536 Cost: 163.92568397521973\n",
            "Epoch: 537 Cost: 163.7205502986908\n",
            "Epoch: 538 Cost: 163.7507827281952\n",
            "Epoch: 539 Cost: 163.72880935668945\n",
            "Epoch: 540 Cost: 163.6518681049347\n",
            "Epoch: 541 Cost: 163.63512098789215\n",
            "Epoch: 542 Cost: 163.71823966503143\n",
            "Epoch: 543 Cost: 163.60410368442535\n",
            "Epoch: 544 Cost: 163.62378323078156\n",
            "Epoch: 545 Cost: 163.64866995811462\n",
            "Epoch: 546 Cost: 163.62604069709778\n",
            "Epoch: 547 Cost: 163.77529454231262\n",
            "Epoch: 548 Cost: 163.68432080745697\n",
            "Epoch: 549 Cost: 163.63236916065216\n",
            "Epoch: 550 Cost: 163.655100107193\n",
            "Epoch: 551 Cost: 163.80904281139374\n",
            "Epoch: 552 Cost: 163.7496622800827\n",
            "Epoch: 553 Cost: 163.75651633739471\n",
            "Epoch: 554 Cost: 163.57673263549805\n",
            "Epoch: 555 Cost: 163.55305576324463\n",
            "Epoch: 556 Cost: 163.5592601299286\n",
            "Epoch: 557 Cost: 163.53616642951965\n",
            "Epoch: 558 Cost: 163.64522171020508\n",
            "Epoch: 559 Cost: 163.5512021780014\n",
            "Epoch: 560 Cost: 163.5665225982666\n",
            "Epoch: 561 Cost: 163.58990466594696\n",
            "Epoch: 562 Cost: 163.5400426387787\n",
            "Epoch: 563 Cost: 163.73390352725983\n",
            "Epoch: 564 Cost: 163.66275572776794\n",
            "Epoch: 565 Cost: 163.67130768299103\n",
            "Epoch: 566 Cost: 163.60612297058105\n",
            "Epoch: 567 Cost: 163.97441792488098\n",
            "Epoch: 568 Cost: 163.48578763008118\n",
            "Epoch: 569 Cost: 163.31647276878357\n",
            "Epoch: 570 Cost: 163.30774676799774\n",
            "Epoch: 571 Cost: 163.4576758146286\n",
            "Epoch: 572 Cost: 163.46303832530975\n",
            "Epoch: 573 Cost: 163.27296364307404\n",
            "Epoch: 574 Cost: 163.35214722156525\n",
            "Epoch: 575 Cost: 163.33814442157745\n",
            "Epoch: 576 Cost: 163.3742127418518\n",
            "Epoch: 577 Cost: 163.3253936767578\n",
            "Epoch: 578 Cost: 163.29543209075928\n",
            "Epoch: 579 Cost: 163.263578414917\n",
            "Epoch: 580 Cost: 163.25072968006134\n",
            "Epoch: 581 Cost: 163.232439994812\n",
            "Epoch: 582 Cost: 163.22598385810852\n",
            "Epoch: 583 Cost: 163.2416272163391\n",
            "Epoch: 584 Cost: 163.21172726154327\n",
            "Epoch: 585 Cost: 163.230464220047\n",
            "Epoch: 586 Cost: 163.19760596752167\n",
            "Epoch: 587 Cost: 163.23398411273956\n",
            "Epoch: 588 Cost: 163.2025614976883\n",
            "Epoch: 589 Cost: 163.2301219701767\n",
            "Epoch: 590 Cost: 163.22171592712402\n",
            "Epoch: 591 Cost: 163.2225260734558\n",
            "Epoch: 592 Cost: 163.20986759662628\n",
            "Epoch: 593 Cost: 163.20214700698853\n",
            "Epoch: 594 Cost: 163.20591390132904\n",
            "Epoch: 595 Cost: 163.19336676597595\n",
            "Epoch: 596 Cost: 163.17372334003448\n",
            "Epoch: 597 Cost: 163.19080567359924\n",
            "Epoch: 598 Cost: 163.16553401947021\n",
            "Epoch: 599 Cost: 163.16614973545074\n",
            "Epoch: 600 Cost: 163.16168582439423\n",
            "Epoch: 601 Cost: 163.15708792209625\n",
            "Epoch: 602 Cost: 163.14454293251038\n",
            "Epoch: 603 Cost: 163.1478111743927\n",
            "Epoch: 604 Cost: 163.1413139104843\n",
            "Epoch: 605 Cost: 163.13694083690643\n",
            "Epoch: 606 Cost: 163.1323035955429\n",
            "Epoch: 607 Cost: 163.12353551387787\n",
            "Epoch: 608 Cost: 163.1296525001526\n",
            "Epoch: 609 Cost: 163.11296224594116\n",
            "Epoch: 610 Cost: 163.13241338729858\n",
            "Epoch: 611 Cost: 163.1188393831253\n",
            "Epoch: 612 Cost: 163.13194465637207\n",
            "Epoch: 613 Cost: 163.19584798812866\n",
            "Epoch: 614 Cost: 163.13296222686768\n",
            "Epoch: 615 Cost: 163.1942012310028\n",
            "Epoch: 616 Cost: 163.10607612133026\n",
            "Epoch: 617 Cost: 163.00539422035217\n",
            "Epoch: 618 Cost: 162.84801697731018\n",
            "Epoch: 619 Cost: 163.11432015895844\n",
            "Epoch: 620 Cost: 163.08012783527374\n",
            "Epoch: 621 Cost: 162.89821863174438\n",
            "Epoch: 622 Cost: 163.03554570674896\n",
            "Epoch: 623 Cost: 162.88469815254211\n",
            "Epoch: 624 Cost: 163.33535766601562\n",
            "Epoch: 625 Cost: 162.4234358072281\n",
            "Epoch: 626 Cost: 162.30778324604034\n",
            "Epoch: 627 Cost: 162.36461329460144\n",
            "Epoch: 628 Cost: 162.09872603416443\n",
            "Epoch: 629 Cost: 161.9979588985443\n",
            "Epoch: 630 Cost: 161.89638090133667\n",
            "Epoch: 631 Cost: 161.8091768026352\n",
            "Epoch: 632 Cost: 161.73506546020508\n",
            "Epoch: 633 Cost: 161.6911417245865\n",
            "Epoch: 634 Cost: 161.68592274188995\n",
            "Epoch: 635 Cost: 161.58635866641998\n",
            "Epoch: 636 Cost: 161.51125013828278\n",
            "Epoch: 637 Cost: 161.48707509040833\n",
            "Epoch: 638 Cost: 161.4614166021347\n",
            "Epoch: 639 Cost: 161.4144401550293\n",
            "Epoch: 640 Cost: 161.3984328508377\n",
            "Epoch: 641 Cost: 161.3041796684265\n",
            "Epoch: 642 Cost: 161.63587844371796\n",
            "Epoch: 643 Cost: 162.8486385345459\n",
            "Epoch: 644 Cost: 163.9835683107376\n",
            "Epoch: 645 Cost: 162.4866954088211\n",
            "Epoch: 646 Cost: 162.91588521003723\n",
            "Epoch: 647 Cost: 160.8307250738144\n",
            "Epoch: 648 Cost: 160.73614704608917\n",
            "Epoch: 649 Cost: 161.02222084999084\n",
            "Epoch: 650 Cost: 160.23690724372864\n",
            "Epoch: 651 Cost: 161.70059072971344\n",
            "Epoch: 652 Cost: 160.7798683643341\n",
            "Epoch: 653 Cost: 160.47323846817017\n",
            "Epoch: 654 Cost: 159.81485641002655\n",
            "Epoch: 655 Cost: 160.46865439414978\n",
            "Epoch: 656 Cost: 160.61450576782227\n",
            "Epoch: 657 Cost: 160.57057082653046\n",
            "Epoch: 658 Cost: 159.51132237911224\n",
            "Epoch: 659 Cost: 159.40364396572113\n",
            "Epoch: 660 Cost: 160.31988751888275\n",
            "Epoch: 661 Cost: 159.49034452438354\n",
            "Epoch: 662 Cost: 159.04112803936005\n",
            "Epoch: 663 Cost: 159.74512028694153\n",
            "Epoch: 664 Cost: 159.12375402450562\n",
            "Epoch: 665 Cost: 158.59127914905548\n",
            "Epoch: 666 Cost: 158.63005113601685\n",
            "Epoch: 667 Cost: 158.08646643161774\n",
            "Epoch: 668 Cost: 159.0551198720932\n",
            "Epoch: 669 Cost: 157.92783272266388\n",
            "Epoch: 670 Cost: 157.85909414291382\n",
            "Epoch: 671 Cost: 157.32243633270264\n",
            "Epoch: 672 Cost: 157.24188256263733\n",
            "Epoch: 673 Cost: 158.12187337875366\n",
            "Epoch: 674 Cost: 157.35911405086517\n",
            "Epoch: 675 Cost: 156.76591658592224\n",
            "Epoch: 676 Cost: 156.98960602283478\n",
            "Epoch: 677 Cost: 156.34954810142517\n",
            "Epoch: 678 Cost: 156.4702546596527\n",
            "Epoch: 679 Cost: 156.0258309841156\n",
            "Epoch: 680 Cost: 155.89633774757385\n",
            "Epoch: 681 Cost: 155.90524184703827\n",
            "Epoch: 682 Cost: 156.068341255188\n",
            "Epoch: 683 Cost: 155.79060304164886\n",
            "Epoch: 684 Cost: 155.86044263839722\n",
            "Epoch: 685 Cost: 155.70378994941711\n",
            "Epoch: 686 Cost: 156.01707935333252\n",
            "Epoch: 687 Cost: 155.51713490486145\n",
            "Epoch: 688 Cost: 155.4214926958084\n",
            "Epoch: 689 Cost: 155.29865682125092\n",
            "Epoch: 690 Cost: 155.33090329170227\n",
            "Epoch: 691 Cost: 155.01894879341125\n",
            "Epoch: 692 Cost: 155.16808772087097\n",
            "Epoch: 693 Cost: 156.39103972911835\n",
            "Epoch: 694 Cost: 155.04029750823975\n",
            "Epoch: 695 Cost: 155.18956065177917\n",
            "Epoch: 696 Cost: 154.99357569217682\n",
            "Epoch: 697 Cost: 155.06577908992767\n",
            "Epoch: 698 Cost: 154.77932631969452\n",
            "Epoch: 699 Cost: 154.65454268455505\n",
            "Epoch: 700 Cost: 154.7366040945053\n",
            "Epoch: 701 Cost: 154.49131166934967\n",
            "Epoch: 702 Cost: 154.46045124530792\n",
            "Epoch: 703 Cost: 154.38522124290466\n",
            "Epoch: 704 Cost: 154.34394991397858\n",
            "Epoch: 705 Cost: 155.4727748632431\n",
            "Epoch: 706 Cost: 154.391455411911\n",
            "Epoch: 707 Cost: 154.39121305942535\n",
            "Epoch: 708 Cost: 156.81618189811707\n",
            "Epoch: 709 Cost: 157.58682477474213\n",
            "Epoch: 710 Cost: 155.66688001155853\n",
            "Epoch: 711 Cost: 160.83013558387756\n",
            "Epoch: 712 Cost: 157.08025693893433\n",
            "Epoch: 713 Cost: 156.54017210006714\n",
            "Epoch: 714 Cost: 156.00928127765656\n",
            "Epoch: 715 Cost: 154.13464844226837\n",
            "Epoch: 716 Cost: 154.538893699646\n",
            "Epoch: 717 Cost: 154.79113447666168\n",
            "Epoch: 718 Cost: 154.29002797603607\n",
            "Epoch: 719 Cost: 154.0111780166626\n",
            "Epoch: 720 Cost: 160.3674796819687\n",
            "Epoch: 721 Cost: 161.0502132177353\n",
            "Epoch: 722 Cost: 156.381693482399\n",
            "Epoch: 723 Cost: 154.25847899913788\n",
            "Epoch: 724 Cost: 153.67718636989594\n",
            "Epoch: 725 Cost: 153.60156869888306\n",
            "Epoch: 726 Cost: 153.47540378570557\n",
            "Epoch: 727 Cost: 154.09310400485992\n",
            "Epoch: 728 Cost: 153.6862336397171\n",
            "Epoch: 729 Cost: 154.8057897090912\n",
            "Epoch: 730 Cost: 154.53978621959686\n",
            "Epoch: 731 Cost: 153.55719709396362\n",
            "Epoch: 732 Cost: 153.26151287555695\n",
            "Epoch: 733 Cost: 153.44175505638123\n",
            "Epoch: 734 Cost: 153.1954609155655\n",
            "Epoch: 735 Cost: 153.29825150966644\n",
            "Epoch: 736 Cost: 153.11288917064667\n",
            "Epoch: 737 Cost: 153.3148353099823\n",
            "Epoch: 738 Cost: 153.19619846343994\n",
            "Epoch: 739 Cost: 152.7221006155014\n",
            "Epoch: 740 Cost: 152.87674963474274\n",
            "Epoch: 741 Cost: 152.9491810798645\n",
            "Epoch: 742 Cost: 153.04467403888702\n",
            "Epoch: 743 Cost: 152.72678768634796\n",
            "Epoch: 744 Cost: 152.90990281105042\n",
            "Epoch: 745 Cost: 152.39583230018616\n",
            "Epoch: 746 Cost: 152.67987501621246\n",
            "Epoch: 747 Cost: 152.22642397880554\n",
            "Epoch: 748 Cost: 152.9573975801468\n",
            "Epoch: 749 Cost: 152.3167107105255\n",
            "Epoch: 750 Cost: 152.20308899879456\n",
            "Epoch: 751 Cost: 151.99387776851654\n",
            "Epoch: 752 Cost: 152.22908902168274\n",
            "Epoch: 753 Cost: 151.99991822242737\n",
            "Epoch: 754 Cost: 151.87335348129272\n",
            "Epoch: 755 Cost: 151.71491837501526\n",
            "Epoch: 756 Cost: 151.5407770872116\n",
            "Epoch: 757 Cost: 151.86653661727905\n",
            "Epoch: 758 Cost: 151.87785804271698\n",
            "Epoch: 759 Cost: 151.48750066757202\n",
            "Epoch: 760 Cost: 151.52627956867218\n",
            "Epoch: 761 Cost: 151.36611998081207\n",
            "Epoch: 762 Cost: 151.2869644165039\n",
            "Epoch: 763 Cost: 151.24350786209106\n",
            "Epoch: 764 Cost: 151.24413752555847\n",
            "Epoch: 765 Cost: 151.1260106563568\n",
            "Epoch: 766 Cost: 151.89842057228088\n",
            "Epoch: 767 Cost: 151.12826085090637\n",
            "Epoch: 768 Cost: 151.61607921123505\n",
            "Epoch: 769 Cost: 151.04036402702332\n",
            "Epoch: 770 Cost: 151.01158738136292\n",
            "Epoch: 771 Cost: 152.3805329799652\n",
            "Epoch: 772 Cost: 151.8783049583435\n",
            "Epoch: 773 Cost: 151.6134910583496\n",
            "Epoch: 774 Cost: 151.72946667671204\n",
            "Epoch: 775 Cost: 152.10601437091827\n",
            "Epoch: 776 Cost: 150.86120569705963\n",
            "Epoch: 777 Cost: 150.9968672990799\n",
            "Epoch: 778 Cost: 150.9485764503479\n",
            "Epoch: 779 Cost: 150.50098550319672\n",
            "Epoch: 780 Cost: 150.39527201652527\n",
            "Epoch: 781 Cost: 150.33895003795624\n",
            "Epoch: 782 Cost: 150.7422102689743\n",
            "Epoch: 783 Cost: 151.31343507766724\n",
            "Epoch: 784 Cost: 151.021901845932\n",
            "Epoch: 785 Cost: 150.70455479621887\n",
            "Epoch: 786 Cost: 150.1366649866104\n",
            "Epoch: 787 Cost: 150.1377693414688\n",
            "Epoch: 788 Cost: 150.16131508350372\n",
            "Epoch: 789 Cost: 149.86083853244781\n",
            "Epoch: 790 Cost: 149.92825973033905\n",
            "Epoch: 791 Cost: 149.9367048740387\n",
            "Epoch: 792 Cost: 150.08405673503876\n",
            "Epoch: 793 Cost: 149.70436775684357\n",
            "Epoch: 794 Cost: 149.75760161876678\n",
            "Epoch: 795 Cost: 149.6965606212616\n",
            "Epoch: 796 Cost: 149.66556060314178\n",
            "Epoch: 797 Cost: 150.07293617725372\n",
            "Epoch: 798 Cost: 149.6741087436676\n",
            "Epoch: 799 Cost: 149.83206295967102\n",
            "Epoch: 800 Cost: 149.92204332351685\n",
            "Epoch: 801 Cost: 149.69692707061768\n",
            "Epoch: 802 Cost: 149.54277396202087\n",
            "Epoch: 803 Cost: 149.57278513908386\n",
            "Epoch: 804 Cost: 149.38426101207733\n",
            "Epoch: 805 Cost: 149.3084877729416\n",
            "Epoch: 806 Cost: 149.2147787809372\n",
            "Epoch: 807 Cost: 149.23161911964417\n",
            "Epoch: 808 Cost: 149.1045536994934\n",
            "Epoch: 809 Cost: 149.07855319976807\n",
            "Epoch: 810 Cost: 149.03291594982147\n",
            "Epoch: 811 Cost: 149.01279389858246\n",
            "Epoch: 812 Cost: 148.99753153324127\n",
            "Epoch: 813 Cost: 148.98458302021027\n",
            "Epoch: 814 Cost: 148.97310590744019\n",
            "Epoch: 815 Cost: 148.9636331796646\n",
            "Epoch: 816 Cost: 148.9550302028656\n",
            "Epoch: 817 Cost: 148.94717860221863\n",
            "Epoch: 818 Cost: 148.9396698474884\n",
            "Epoch: 819 Cost: 148.93251526355743\n",
            "Epoch: 820 Cost: 148.9247601032257\n",
            "Epoch: 821 Cost: 148.9173732995987\n",
            "Epoch: 822 Cost: 148.9110666513443\n",
            "Epoch: 823 Cost: 148.9064667224884\n",
            "Epoch: 824 Cost: 148.90251553058624\n",
            "Epoch: 825 Cost: 148.89706826210022\n",
            "Epoch: 826 Cost: 148.8833680152893\n",
            "Epoch: 827 Cost: 148.89025366306305\n",
            "Epoch: 828 Cost: 148.89909493923187\n",
            "Epoch: 829 Cost: 148.7984254360199\n",
            "Epoch: 830 Cost: 148.72696447372437\n",
            "Epoch: 831 Cost: 148.71275389194489\n",
            "Epoch: 832 Cost: 148.69556307792664\n",
            "Epoch: 833 Cost: 148.68705832958221\n",
            "Epoch: 834 Cost: 148.67924284934998\n",
            "Epoch: 835 Cost: 148.67247366905212\n",
            "Epoch: 836 Cost: 148.66657948493958\n",
            "Epoch: 837 Cost: 148.66045999526978\n",
            "Epoch: 838 Cost: 148.65519392490387\n",
            "Epoch: 839 Cost: 148.64971709251404\n",
            "Epoch: 840 Cost: 148.64426815509796\n",
            "Epoch: 841 Cost: 148.63954424858093\n",
            "Epoch: 842 Cost: 148.63465106487274\n",
            "Epoch: 843 Cost: 148.63018703460693\n",
            "Epoch: 844 Cost: 148.62431395053864\n",
            "Epoch: 845 Cost: 148.62329506874084\n",
            "Epoch: 846 Cost: 148.61775588989258\n",
            "Epoch: 847 Cost: 148.60312795639038\n",
            "Epoch: 848 Cost: 148.58605480194092\n",
            "Epoch: 849 Cost: 148.562109708786\n",
            "Epoch: 850 Cost: 148.55199921131134\n",
            "Epoch: 851 Cost: 148.53984797000885\n",
            "Epoch: 852 Cost: 148.5461641550064\n",
            "Epoch: 853 Cost: 148.6282877922058\n",
            "Epoch: 854 Cost: 148.47696912288666\n",
            "Epoch: 855 Cost: 148.45958960056305\n",
            "Epoch: 856 Cost: 148.4482043981552\n",
            "Epoch: 857 Cost: 148.4413069486618\n",
            "Epoch: 858 Cost: 148.43448841571808\n",
            "Epoch: 859 Cost: 148.42934787273407\n",
            "Epoch: 860 Cost: 148.42473566532135\n",
            "Epoch: 861 Cost: 148.42070722579956\n",
            "Epoch: 862 Cost: 148.4174027442932\n",
            "Epoch: 863 Cost: 148.41424810886383\n",
            "Epoch: 864 Cost: 148.4109845161438\n",
            "Epoch: 865 Cost: 148.40763092041016\n",
            "Epoch: 866 Cost: 148.4049233198166\n",
            "Epoch: 867 Cost: 148.40232455730438\n",
            "Epoch: 868 Cost: 148.39933919906616\n",
            "Epoch: 869 Cost: 148.39655923843384\n",
            "Epoch: 870 Cost: 148.394673705101\n",
            "Epoch: 871 Cost: 148.39178895950317\n",
            "Epoch: 872 Cost: 148.38979256153107\n",
            "Epoch: 873 Cost: 148.38751769065857\n",
            "Epoch: 874 Cost: 148.3856316804886\n",
            "Epoch: 875 Cost: 148.38386380672455\n",
            "Epoch: 876 Cost: 148.38183546066284\n",
            "Epoch: 877 Cost: 148.3798645734787\n",
            "Epoch: 878 Cost: 148.37794387340546\n",
            "Epoch: 879 Cost: 148.37638366222382\n",
            "Epoch: 880 Cost: 148.37480247020721\n",
            "Epoch: 881 Cost: 148.373086810112\n",
            "Epoch: 882 Cost: 148.37167692184448\n",
            "Epoch: 883 Cost: 148.37003803253174\n",
            "Epoch: 884 Cost: 148.3686157464981\n",
            "Epoch: 885 Cost: 148.36968064308167\n",
            "Epoch: 886 Cost: 148.80829417705536\n",
            "Epoch: 887 Cost: 148.33012008666992\n",
            "Epoch: 888 Cost: 148.28968715667725\n",
            "Epoch: 889 Cost: 148.34897792339325\n",
            "Epoch: 890 Cost: 148.23641169071198\n",
            "Epoch: 891 Cost: 148.21448493003845\n",
            "Epoch: 892 Cost: 148.14845323562622\n",
            "Epoch: 893 Cost: 148.11131942272186\n",
            "Epoch: 894 Cost: 148.1002641916275\n",
            "Epoch: 895 Cost: 148.09266269207\n",
            "Epoch: 896 Cost: 148.0871285200119\n",
            "Epoch: 897 Cost: 148.0824762582779\n",
            "Epoch: 898 Cost: 148.07865405082703\n",
            "Epoch: 899 Cost: 148.07550156116486\n",
            "Epoch: 900 Cost: 148.07274675369263\n",
            "Epoch: 901 Cost: 148.0706284046173\n",
            "Epoch: 902 Cost: 148.06846678256989\n",
            "Epoch: 903 Cost: 148.06658971309662\n",
            "Epoch: 904 Cost: 148.0646722316742\n",
            "Epoch: 905 Cost: 148.06319546699524\n",
            "Epoch: 906 Cost: 148.06164503097534\n",
            "Epoch: 907 Cost: 148.05990195274353\n",
            "Epoch: 908 Cost: 148.05868518352509\n",
            "Epoch: 909 Cost: 148.05722653865814\n",
            "Epoch: 910 Cost: 148.05596792697906\n",
            "Epoch: 911 Cost: 148.0546886920929\n",
            "Epoch: 912 Cost: 148.05344700813293\n",
            "Epoch: 913 Cost: 148.05238711833954\n",
            "Epoch: 914 Cost: 148.05145692825317\n",
            "Epoch: 915 Cost: 148.05002391338348\n",
            "Epoch: 916 Cost: 148.0490471124649\n",
            "Epoch: 917 Cost: 148.04786157608032\n",
            "Epoch: 918 Cost: 148.04679656028748\n",
            "Epoch: 919 Cost: 148.04559099674225\n",
            "Epoch: 920 Cost: 148.0446399450302\n",
            "Epoch: 921 Cost: 148.0435460805893\n",
            "Epoch: 922 Cost: 148.04265654087067\n",
            "Epoch: 923 Cost: 148.04153096675873\n",
            "Epoch: 924 Cost: 148.04060876369476\n",
            "Epoch: 925 Cost: 148.03955125808716\n",
            "Epoch: 926 Cost: 148.03863883018494\n",
            "Epoch: 927 Cost: 148.03767156600952\n",
            "Epoch: 928 Cost: 148.03670156002045\n",
            "Epoch: 929 Cost: 148.03576004505157\n",
            "Epoch: 930 Cost: 148.03480434417725\n",
            "Epoch: 931 Cost: 148.0337975025177\n",
            "Epoch: 932 Cost: 148.0327980518341\n",
            "Epoch: 933 Cost: 148.0316503047943\n",
            "Epoch: 934 Cost: 148.03072452545166\n",
            "Epoch: 935 Cost: 148.02952921390533\n",
            "Epoch: 936 Cost: 148.02843570709229\n",
            "Epoch: 937 Cost: 148.02721214294434\n",
            "Epoch: 938 Cost: 148.02602458000183\n",
            "Epoch: 939 Cost: 148.0247619152069\n",
            "Epoch: 940 Cost: 148.0236518383026\n",
            "Epoch: 941 Cost: 148.02262198925018\n",
            "Epoch: 942 Cost: 148.02177560329437\n",
            "Epoch: 943 Cost: 148.0205249786377\n",
            "Epoch: 944 Cost: 148.0193704366684\n",
            "Epoch: 945 Cost: 148.01831150054932\n",
            "Epoch: 946 Cost: 148.01711511611938\n",
            "Epoch: 947 Cost: 148.01609361171722\n",
            "Epoch: 948 Cost: 148.0162124633789\n",
            "Epoch: 949 Cost: 148.02088510990143\n",
            "Epoch: 950 Cost: 148.04050600528717\n",
            "Epoch: 951 Cost: 147.96647548675537\n",
            "Epoch: 952 Cost: 147.943012714386\n",
            "Epoch: 953 Cost: 147.93369925022125\n",
            "Epoch: 954 Cost: 147.92878305912018\n",
            "Epoch: 955 Cost: 147.92511212825775\n",
            "Epoch: 956 Cost: 147.92213833332062\n",
            "Epoch: 957 Cost: 147.9198487997055\n",
            "Epoch: 958 Cost: 147.91787457466125\n",
            "Epoch: 959 Cost: 147.91594111919403\n",
            "Epoch: 960 Cost: 147.9142061471939\n",
            "Epoch: 961 Cost: 147.9124435186386\n",
            "Epoch: 962 Cost: 147.9107723236084\n",
            "Epoch: 963 Cost: 147.90965282917023\n",
            "Epoch: 964 Cost: 147.9093737602234\n",
            "Epoch: 965 Cost: 147.91303133964539\n",
            "Epoch: 966 Cost: 148.0178552865982\n",
            "Epoch: 967 Cost: 147.95040941238403\n",
            "Epoch: 968 Cost: 147.8778793811798\n",
            "Epoch: 969 Cost: 147.84311962127686\n",
            "Epoch: 970 Cost: 147.82760751247406\n",
            "Epoch: 971 Cost: 147.82367384433746\n",
            "Epoch: 972 Cost: 147.82045531272888\n",
            "Epoch: 973 Cost: 147.8180959224701\n",
            "Epoch: 974 Cost: 147.8157660961151\n",
            "Epoch: 975 Cost: 147.81396520137787\n",
            "Epoch: 976 Cost: 147.81219065189362\n",
            "Epoch: 977 Cost: 147.81062400341034\n",
            "Epoch: 978 Cost: 147.80904471874237\n",
            "Epoch: 979 Cost: 147.8076937198639\n",
            "Epoch: 980 Cost: 147.80611050128937\n",
            "Epoch: 981 Cost: 147.8051471710205\n",
            "Epoch: 982 Cost: 147.8040030002594\n",
            "Epoch: 983 Cost: 147.8031198978424\n",
            "Epoch: 984 Cost: 147.8021184206009\n",
            "Epoch: 985 Cost: 147.80122017860413\n",
            "Epoch: 986 Cost: 147.80028474330902\n",
            "Epoch: 987 Cost: 147.79939353466034\n",
            "Epoch: 988 Cost: 147.79855501651764\n",
            "Epoch: 989 Cost: 147.7977877855301\n",
            "Epoch: 990 Cost: 147.79699170589447\n",
            "Epoch: 991 Cost: 147.7960685491562\n",
            "Epoch: 992 Cost: 147.795339345932\n",
            "Epoch: 993 Cost: 147.7945384979248\n",
            "Epoch: 994 Cost: 147.79366743564606\n",
            "Epoch: 995 Cost: 147.79282474517822\n",
            "Epoch: 996 Cost: 147.79208743572235\n",
            "Epoch: 997 Cost: 147.79119789600372\n",
            "Epoch: 998 Cost: 147.79040718078613\n",
            "Epoch: 999 Cost: 147.78968465328217\n"
          ]
        }
      ],
      "source": [
        "## Initial values for cost tracking\n",
        "track_cost = np.zeros(epochs)\n",
        "cur_cost = 0.0\n",
        "\n",
        "## Loop through the data\n",
        "for epoch in range(epochs):\n",
        "    \n",
        "    cur_cost = 0.0\n",
        "    correct = 0.0\n",
        "    \n",
        "    ## train_loader is iterable and numbers knows the batch\n",
        "    for i, data in enumerate(train_loader, 0):\n",
        "        \n",
        "        ## The input tensor and labels tensor for the current batch\n",
        "        inputs, labels = data\n",
        "        \n",
        "        ## Clear the gradient from the previous batch\n",
        "        optimizer.zero_grad()\n",
        "        \n",
        "        ## Provide the input tensor into the network to get outputs\n",
        "        outputs = net(inputs)\n",
        "\n",
        "        ## Calculate the cost for the current batch\n",
        "        ## nn.Softmax is used because net outputs prediction scores and our cost function expects probabilities and labels\n",
        "        cost = cost_fn(nn.Softmax(dim=1)(outputs), labels)\n",
        "        \n",
        "        ## Calculate the gradient\n",
        "        cost.backward()\n",
        "        \n",
        "        ## Update the model parameters using the gradient\n",
        "        optimizer.step()\n",
        "        \n",
        "        ## Track the current cost (accumulating across batches)\n",
        "        cur_cost += cost.item()\n",
        "    \n",
        "    ## Store the accumulated cost at each epoch\n",
        "    track_cost[epoch] = cur_cost\n",
        "    print(f\"Epoch: {epoch} Cost: {cur_cost}\") ## Uncomment this if you want printed updates"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 151,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "### Plot track costs\n",
        "plt.plot(np.linspace(0, epochs, epochs), track_cost)\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 152,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "0.22857142857142856\n"
          ]
        }
      ],
      "source": [
        "## Initialize objects for counting correct/total \n",
        "correct = 0\n",
        "total = 0\n",
        "\n",
        "### Convert data frame to Tensor for DataLoader\n",
        "y_test_tensor = torch.Tensor(test_y)\n",
        "X_test_tensor = torch.from_numpy(test_X)\n",
        "test_loader = DataLoader(TensorDataset(X_test_tensor.type(torch.FloatTensor), \n",
        "                        y_test_tensor.type(torch.LongTensor)), batch_size=bsize)\n",
        "\n",
        "### Calculate accuracy for test set\n",
        "with torch.no_grad():\n",
        "    for data in test_loader:\n",
        "        images, labels = data\n",
        "        outputs = net(images)\n",
        "        _, predicted = torch.max(outputs.data, 1)\n",
        "        total += labels.size(0)\n",
        "        correct += (predicted == labels).sum().item()\n",
        "        \n",
        "print(correct/total)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 153,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "0.4326530612244898\n"
          ]
        }
      ],
      "source": [
        "## Initialize objects for counting correct/total \n",
        "correct = 0\n",
        "total = 0\n",
        "n_predictions = 3\n",
        "\n",
        "### Calculate accuracy for test set using the first 3 predictions\n",
        "with torch.no_grad():\n",
        "    for data in test_loader:\n",
        "        images, labels = data\n",
        "        outputs = net(images)\n",
        "        _, topi = outputs.topk(n_predictions, 1, True)\n",
        "        for i in range(topi.size(0)):\n",
        "            prediction_set = set(topi[i].tolist())\n",
        "            correct += (labels[i].item() in prediction_set)\n",
        "        total += labels.size(0)\n",
        "        \n",
        "print(correct/total)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 154,
      "metadata": {},
      "outputs": [],
      "source": [
        "### Helper function to generate one melspectrogram for an mp3 file\n",
        "def generate_one_spectrogram(input_filename):\n",
        "    TEST_PATH = 'test_files/'\n",
        "\n",
        "    ### Generate spectrogram\n",
        "    src = TEST_PATH + input_filename + '.mp3'\n",
        "    dst = TEST_PATH + input_filename + '.wav'\n",
        "                                 \n",
        "    sound = AudioSegment.from_mp3(src)\n",
        "    sound.export(dst, format=\"wav\")\n",
        "\n",
        "    SAMPLE_RATE = 22050\n",
        "    y, sr = librosa.load(dst,sr=SAMPLE_RATE)\n",
        "            \n",
        "    S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128,\n",
        "                            fmax=8000)\n",
        "    fig, ax = plt.subplots()\n",
        "    S_dB = librosa.power_to_db(S, ref=np.max)\n",
        "    img = librosa.display.specshow(S_dB, x_axis='time',\n",
        "                            y_axis='mel', sr=sr,\n",
        "                            fmax=8000, ax=ax)\n",
        "    \n",
        "    plt.axis('off')\n",
        "    plt.savefig(TEST_PATH + input_filename + \".png\", bbox_inches='tight', pad_inches=0)\n",
        "    return TEST_PATH + input_filename + \".png\"\n",
        "\n",
        "### Helper function to convert a melspectrogram image into a tensor object\n",
        "def convert_img_to_tensor(img_filename):\n",
        "    image = plt.imread(img_filename)\n",
        "\n",
        "    ## Convert to tensor and check the shape\n",
        "    spec = torch.from_numpy(image)\n",
        "\n",
        "    ## Move third dimension (color channels) to the first dimension\n",
        "    spec2 = torch.movedim(spec, source=2, destination=0)\n",
        "\n",
        "    ## Add an empty first dimension, putting our tensor into the standard format\n",
        "    spec_final = torch.unsqueeze(spec2, dim=0)\n",
        "    \n",
        "    spec_resized = functional.interpolate(spec_final, size = (img_height,img_width))\n",
        "    \n",
        "    ## spec_img_format = torch.movedim(spec_resized[0], source=0, destination=2)\n",
        "\n",
        "    print(spec_resized.shape)\n",
        "    \n",
        "    return spec_resized\n",
        "\n",
        "### Array of genre mappings to help print predictions\n",
        "CATEGORY_LABEL = [\"acoustic\", \"blues\", \"classical\", \"country\", \"dance\", \"edm\", \"hip-hop\", \"indie\", \"jazz\", \"pop\", \"rock\", \"world-music\"]\n",
        "\n",
        "### Take in an mp3 filename to display k predictions and their scores\n",
        "def predict(input_filename, n_predictions=6):\n",
        "    print('\\n> %s' % input_filename)\n",
        "\n",
        "    spectrogram_filename = generate_one_spectrogram(input_filename)\n",
        "    img_tensor = convert_img_to_tensor(spectrogram_filename)\n",
        "    \n",
        "    ## Don't update gradient with any of these examples\n",
        "    with torch.no_grad():\n",
        "        ## Find top predictions \n",
        "        output = net(img_tensor)\n",
        "        topv, topi = output.topk(n_predictions, 1, True)\n",
        "        predictions = []\n",
        "\n",
        "        ## Actual label\n",
        "        actual_category = input_filename.split('_')[0]\n",
        "        print(f\"\\nActual Genre: {actual_category}\\n\")\n",
        "        \n",
        "        ## Go through the category predictions and save info for printing\n",
        "        print(\"Top Predicted Genres:\")\n",
        "        for i in range(n_predictions):\n",
        "            value = topv[0][i].item()\n",
        "            category_index = topi[0][i].item()\n",
        "            print('(%.2f) %s' % (value, CATEGORY_LABEL[category_index]))\n",
        "            predictions.append([value, CATEGORY_LABEL[category_index]])"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.11.2"
    },
    "orig_nbformat": 4
  },
  "nbformat": 4,
  "nbformat_minor": 2
}