BERT.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "toc_visible": true,
      "machine_shape": "hm",
      "gpuType": "V100"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "## **We attempt to use BERT for multi-class classification of toxic texts.**"
      ],
      "metadata": {
        "id": "eqPqTFIW1Xn_"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "\n",
        "import torch\n",
        "from torch.utils.data import DataLoader, TensorDataset\n",
        "from transformers import BertTokenizer, BertForSequenceClassification, AdamW\n",
        "from sklearn.model_selection import train_test_split\n",
        "import torch\n",
        "from torch import nn\n",
        "from torch.utils.data import Dataset, DataLoader\n",
        "from transformers import BertTokenizer, BertModel\n",
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "\n",
        "import transformers\n",
        "from transformers import AdamW, get_linear_schedule_with_warmup\n",
        "from tqdm import tqdm\n",
        "import seaborn as sns\n",
        "import matplotlib as plt\n",
        "import torch.nn.functional as Func"
      ],
      "metadata": {
        "id": "XDjTNciSPnQ6"
      },
      "execution_count": 40,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "These are all the imports, initially we made use of the scheduler but later deicded against the added complexity of code and insignificant performance gains. Other imports are primarily used for functions tasks."
      ],
      "metadata": {
        "id": "itlthcUi1GUf"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "\n",
        "# Load your dataset\n",
        "file_path = '/content/train.csv'  # Replace with the path to your dataset file\n",
        "data = pd.read_csv(file_path)\n",
        "\n",
        "# Initialize the BERT tokenizer\n",
        "tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\n",
        "\n",
        "# Tokenize and count tokens for each text\n",
        "token_counts = []\n",
        "for text in data['comment_text']:\n",
        "    # Encode the text, adding special tokens for BERT (CLS and SEP)\n",
        "    encoded_text = tokenizer.encode(text, add_special_tokens=True)\n",
        "    token_counts.append(len(encoded_text))\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "o49DaO0ZZUvg",
        "outputId": "0a552b90-9ee9-462f-c6b6-551bfa9989df"
      },
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Token indices sequence length is longer than the specified maximum sequence length for this model (631 > 512). Running this sequence through the model will result in indexing errors\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\n",
        "token_counts = [len(tokenizer.encode(text, add_special_tokens=True)) for text in data['comment_text']]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ZTCm9wVUWjcL",
        "outputId": "f5aa5b8d-358a-4b3d-9967-97e8e2e259b1"
      },
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Token indices sequence length is longer than the specified maximum sequence length for this model (631 > 512). Running this sequence through the model will result in indexing errors\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Create a histogram of token counts\n",
        "import matplotlib.pyplot as plt\n",
        "plt.figure(figsize=(10, 6))\n",
        "sns.histplot(token_counts, bins=100, kde=False)\n",
        "plt.title('Distribution of Token Counts in Texts')\n",
        "plt.xlabel('Token Count')\n",
        "plt.ylabel('Number of Texts')\n",
        "sns.set_style(\"white\")  # Set style to white for better appearance in LaTeX\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 564
        },
        "id": "_mesi340Vw3H",
        "outputId": "778beaf4-061b-4f9e-e16c-0a7bb85e0ddf"
      },
      "execution_count": 42,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1000x600 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA2QAAAIjCAYAAABswtioAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAABf6ElEQVR4nO3df3xP9f//8fs2W36MYSNvPwpbNsx+oLe2XqwWEVGz/P6dovyqd8TIu5AMlTQUIT9mQvlRQr+EsE3R1iK/J9HesU3M/Nps5/tHX+fj1ZY2bQ52u14uu1y8zvNxznmc1w523znn+XIwDMMQAAAAAOCGc7S6AQAAAAAoqQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAUEgzZsyQt7f3DdlXr1691KtXL/P1jh075O3trc8+++yG7D8iIkKhoaE3ZF/X69y5c3rppZd0//33y9vbW6+99lqx7u/K9//UqVPFuh9c2/Hjx+Xt7a1Vq1ZZ3QoA/COlrG4AAKy0atUqjR492nzt4uIiNzc3eXt7KyQkRB07dpSrq+s/3s+JEye0YsUKtWzZUvXr1//H2ytKN3NvBTFnzhytXr1agwYNUq1ateTp6ZmnZsaMGZo5c+bfbuvf//63oqOji6PNYpWWlqb58+dr06ZN+t///icHBwfVrVtXLVu2VM+ePVWhQgWrW9TatWuVnp6uvn37Wt1KHn/+d+Cv1KhRQ19//XWR7HPLli1KSkrS0KFDi2R7AG5dBDIAkDRs2DDVrFlTly9fVlpamr799ltNmjRJCxcu1DvvvCMfHx+z9tlnn9WAAQMKtf2TJ09q5syZqlGjRqFCz/z58wu1n+txrd5effVVGYZR7D38E/Hx8fL399eQIUP+sqZVq1a66667zNfnz5/XuHHj1KpVK7Vq1cpc7uHhUay9FoekpCQNGDBA58+fV4cOHdSwYUNJ0u7duzV37lzt3LlT77//vsVdSp9++qkOHjxYZIGsRo0aSkpKUqlS//xHmXvvvVdTp061WzZ27Fj5+fmpc+fO5rJy5cr9431dsWXLFsXExBDIABDIAECSWrRooUaNGpmvBw4cqLi4OD3zzDMaNGiQ1q9fr9KlS0uSSpUqVSQ/BF7LhQsXVKZMGbm4uBTrfv6Os7OzpfsviPT0dHl5eV2zxsfHxy5Unzp1SuPGjZO3t7cee+yx4m6x2GRkZGjIkCFycnLS6tWr81wd/M9//qMVK1ZY1F3xcnBw0B133FEk26pVq5Zq1aplt2zcuHGqVavWLX1+ALg18AwZAPyFoKAgDRo0SL/++qs++eQTc3l+z5Bt375d3bp1U9OmTRUYGKjWrVtr2rRpkv547uuJJ56QJI0ePVre3t52z7706tVLjz76qHbv3q0ePXrI39/fXPfPz5BdkZubq2nTpun+++9XQECAnnnmGf3vf/+zqwkNDVVERESeda/e5t/1lt8zZOfPn9fkyZMVEhIiX19ftW7dWvPnz89zJc3b21sTJkzQV199pUcffVS+vr5q166dvvnmm2u97ab09HSNGTNGwcHBatSokTp06KDVq1eb41eepzt+/Lg2b95s9n78+PECbT8/cXFx6t69uwICAtS0aVM9++yzOnz48N+u9+uvv6pVq1Z69NFHlZaWJumPsPTaa6+Z71OrVq303nvvKTc311zvynNQ8+fP1/Lly9WyZUv5+voqPDxcSUlJf7vfZcuW6cSJE4qIiMj3Vk0PDw8NGjTIbllMTIzatWsnX19f2Ww2jR8/XhkZGXY1BTl3pP/7Hqxfv17vvvuu+YuNPn366OjRo3brbd68Wb/++qv5fbr6vIqOjla7du3k7++ve++9Vx07dtTatWuveez5PUMWERGhwMBAnThxQoMGDVJgYKDuu+8+TZkyRTk5OdfcXkGcOHFCo0ePVnBwsHk+f/TRR+b4xYsX1aZNG7Vp00YXL140l58+fVo2m01du3ZVTk6OIiIiFBMTI0nm+3H1vynr1q1Tx44dFRgYqMaNG6t9+/ZatGjRP+4fwM2JK2QAcA2PPfaYpk2bpm3bttndunS1gwcPauDAgfL29tawYcPk4uKio0eP6vvvv5ckeXp6atiwYYqKilKXLl3UpEkTSVLjxo3NbZw+fVpPP/202rVrpw4dOsjd3f2afb377rtycHDQ008/rfT0dC1atEh9+/bVxx9/bF7JK4iC9HY1wzD07LPPmkGufv362rp1q6ZOnaoTJ05ozJgxdvW7du3SF198oe7du6tcuXKKjo7WsGHDtGnTJlWqVOkv+7p48aJ69eqlX375RT169FDNmjX12WefKSIiQhkZGerTp488PT01depURUZGqlq1aurXr58kqXLlygU+/qvFxsbq6aefVs2aNTVkyBBdvHhRS5YsUbdu3bRq1SrVrFkz3/V++eUX9enTR25ubnr//fdVuXJlXbhwQT179tSJEyfUtWtX/etf/1JCQoKmTZum1NRUvfTSS3bb+PTTT3Xu3Dl16dJFDg4OmjdvnoYOHaqvvvrqmlcpv/76a5UuXVqtW7cu0DFeeZYuODhY3bp105EjR/TBBx/oxx9/1AcffHDdV0Tnzp0rBwcHPfnkk8rMzNS8efM0YsQIffjhh5KkZ555RmfPntVvv/1mPqt15fa/FStWaOLEiWrdurV69+6tS5cuaf/+/frhhx/Uvn37QveSk5Oj/v37y8/PTyNHjlRcXJzef/991apVS927d7+u45P+eE6vc+fOcnBwUI8ePVS5cmV98803eumll5SZmam+ffuqdOnSmjJlirp166a33nrLPNYJEybo7NmzioyMlJOTk7p06aKTJ09q+/bteW6V3L59u1544QUFBQVpxIgRkqTk5GR9//336tOnz3X3D+DmRSADgGuoVq2aypcvr2PHjv1lzfbt25Wdna25c+fmGwY8PDzUokULRUVFKSAgIN9boFJTUzV+/Hh17dq1QH2dOXNG69evNyccadCggZ5//nmtWLFCvXv3LuDRFay3q23cuFHx8fF6/vnn9eyzz0qSevTooWHDhmnx4sXq2bOn3bNahw8f1vr1681lzZo102OPPaZ169apZ8+ef7mf5cuX6/Dhw3r99dfVoUMHSVLXrl3Vq1cvTZ8+XeHh4fLw8NBjjz2mt99+W3feeec/vrVs6tSpcnNz0/Lly1WxYkVJUsuWLRUWFqYZM2ZoypQpedY5fPiw+vbtqzvvvFPz58+Xm5ubJGnBggU6duyYVq9erdq1a5v9V61aVfPnz9eTTz6pf/3rX+Z2UlJS9MUXX5jr16lTR4MGDdK2bdv04IMP/mXPycnJql27doFubT116pTmzJkjm82muXPnytHxj5tk6tatqwkTJuiTTz5ReHh4gd6rP7t06ZLWrFlj9lGhQgW99tprOnDggOrVq6f7779fixcvVkZGRp7v0+bNm3XPPfcoKirquvadXy+PPPKIBg8eLEnq1q2bwsLC9NFHH/2jQPbWW28pJydHa9euNX+Z0K1bN73wwguaOXOmunbtqtKlS8vf319PPfWU5s6dq1atWiktLU3r1q3TmDFjVKdOHUlSYGCgateure3bt+f7fri6umr+/PlycnK67n4B3Dq4ZREA/kbZsmV17ty5vxy/MoPdxo0b7W5HKwwXFxd17NixwPWPP/643eyPbdq0UZUqVbRly5br2n9BffPNN3JycspzG+WTTz4pwzDy3I4YHBxsF9B8fHzk6up6zYB7ZT9VqlTRo48+ai5zdnZWr169dP78eX333XdFcDT/5+TJk9q7d6/CwsLMMHal3+Dg4Hzf14MHD6pXr16qUaOGFi5caIYpSfrss8/UpEkTVahQQadOnTK/goODlZOTk6f/tm3b2q3ftGlTSfrb9ykzM7PAE03ExsYqOztbvXv3NsOYJHXq1Emurq7/6Nzp2LGjXSgsaP/SH39/fvvttwLdollQ3bp1s3vdpEmTf3Qrq2EY+uKLLxQaGirDMOy+pzabTWfPntWePXvM+iFDhsjLy0ujRo3S+PHj9e9//7vAvyipUKGCLly4oO3bt193vwBuLVwhA4C/cf78+WveQti2bVt9+OGHGjt2rN58800FBQWpVatWatOmjd0Pvtdy5513FmoCj7vvvtvutYODg+6++279+uuvBd7G9fj1119VtWrVPB8FcOX5pT/v/+qrQFe4ubnleWYpv/3cfffded6/K/tJSUkpdO/XcmV7V65g/Hmf27Zt0/nz51W2bFlz+TPPPCMPDw/Nnz8/Tyg6evSo9u/fr6CgoHz39+fPMPvz+3QlnP3d++Tq6nrNXxZc7cox1q1b1265i4uLatWq9Y/OnerVq9u9vvJLir/rX5KefvppxcbGqlOnTrr77rt1//3369FHHzVvny2sO+64I8+Vajc3N505c+a6tif98f3KyMjQ8uXLtXz58r+sucLFxUWTJk3SE088oTvuuEOTJk2Sg4NDgfbVvXt3bdiwQU8//bTuvPNO3X///XrkkUfUokWL6+4fwM2NQAYA1/Dbb7/p7Nmzdld5/qx06dKKiYnRjh07tHnzZm3dulXr16/X8uXL9f777xfotqPCPPf1T+Xk5NywW6H+aj83+1T6BdG6dWutXr1aa9euzXOraW5uru6//3499dRT+a575TbGK673fapbt6727t2rrKysGzIj51+dO3/1i4eCfJ89PT312WefmX93vvjiCy1dulSDBw/WsGHDCt1jcZzbV658d+jQQWFhYfnW/Hmin23btkn64xbKo0eP5pnF8a+4u7trzZo12rZtm7755ht98803WrVqlR5//PF8b5sFcOsjkAHANXz88ceSJJvNds06R0dHBQUFKSgoSKNHj9bs2bP11ltvaceOHQoODi7wb8cL6uoZ7KQ/fvA9evSo3Q+Ff3UlKiUlxe6Hw8L0VqNGDcXFxSkzM9PuKllycrI5XhRq1Kih/fv3Kzc31+6H/Sv7+fMVmX/qyvaOHDmSZyw5OVmVKlWyuzomSSNHjpSTk5PGjx+vcuXK2U1Acdddd+n8+fMKDg4u0j7/7MEHH1RCQoK++OILu9s783PlGJOTk+2+/1lZWTp+/LhdrwU9dwrjWudZ2bJl1bZtW7Vt21ZZWVkaOnSoZs+erYEDBxbZ1Pb/ROXKlVWuXDnl5uYW6Hu6b98+zZo1Sx07dtS+ffs0duxYrV27VuXLlzdrrvV+uLi4KDQ0VKGhocrNzdW4ceO0fPlyDRo0KM/VcQC3Pp4hA4C/EBcXp3feeUc1a9Y0J5bIz+nTp/Msu/IBy1lZWZKkMmXKSCrYLVwFsWbNGmVmZpqvP/vsM6Wmptrd1lSrVi398MMPZg+StGnTpjzT4xemtxYtWignJ8ecsvuKhQsXysHBochuq2rRooVSU1O1fv16c9nly5cVHR2tsmXL6t577y2S/VxRtWpV1a9fX2vWrLF7Hw4cOKDt27crJCQk3/VeffVVtW7dWhEREdq4caO5/JFHHlFCQoK2bt2aZ52MjAxdvny5SPru2rWrqlSposmTJ+cbJtPT0/XOO+9I+uN5PmdnZ0VHR9tdufroo4909uxZu2Ms6LlTGGXKlNHZs2fzLP/999/tXru4uMjT01OGYSg7O/u691eUnJyc1Lp1a33++ec6cOBAnvGrb1fMzs7W6NGjVbVqVb300kuKjIxUWlqaJk2aZLfOX/29+/P74ejoaP6i5ervB4DbB1fIAEB/TCKRnJysnJwcpaWlaceOHdq+fbuqV6+ud99995q/pZ81a5Z27typkJAQ1ahRQ+np6Vq6dKmqVatmPgdz1113qUKFClq2bJnKlSunsmXLys/P77qvNri5ual79+7q2LGjOe393XffbTc1f6dOnfT555/rqaee0iOPPKJffvlFa9euzXP7ZWF6Cw0NVbNmzfTWW2+Znym1fft2bdy4UX369LnmrZ2F0aVLFy1fvlwRERHas2ePatSooc8//1zff/+9xowZk+cZtqIwcuRIPf300+rSpYueeOIJc9r78uXLa8iQIfmu4+joqNdff12DBw/W888/r/fee09BQUHq37+/vv76az3zzDMKCwtTw4YNdeHCBR04cECff/65Nm7ceN3T81/Nzc1Ns2bN0oABA/T444+rQ4cOatiwoSTpp59+0qeffqrAwEBJf1zlGThwoGbOnKmnnnpKoaGhOnLkiJYuXWp+ztsVBT13CqNhw4Zav369IiMj1ahRI5UtW1ahoaHq37+/PDw81LhxY7m7uys5OVlLlixRSEhIsXyfr9fw4cO1Y8cOde7cWZ06dZKXl5fOnDmjPXv2KC4uTt9++62kPz6SYu/evVq4cKFcXV3l4+OjwYMHa/r06WrTpo0ZfK98nyZOnCibzSYnJye1a9dOY8eO1ZkzZ3TffffpzjvvVEpKipYsWaL69evn+1lzAG59BDIAkMwpt52dnVWxYkXVq1dPY8aMUceOHf/2h8LQ0FD9+uuvWrlypX7//XdVqlRJ//73vzV06FDzFiVnZ2dNnjxZ06ZN07hx43T58mVFRkZedyB75plntH//fr333ns6d+6cgoKC9Morr5i/dZek5s2bKyIiQgsWLNCkSZPk6+ur2bNn53kOpTC9OTo66t1331VUVJTWr1+vVatWqUaNGho5cqSefPLJ6zqW/JQuXVrR0dF64403tHr1amVmZqpOnTqKjIws1GyUhREcHKx58+YpKipKUVFRKlWqlO699169+OKL1/w+OTs7KyoqSk8//bQGDRqkhQsXyt/fX9HR0ZozZ44+++wzrVmzRq6urqpdu7bdeVEU/P39tXbtWs2fP1+bN2/Wxx9/LEdHR9WtW1cDBgyw+3iBoUOHqnLlylqyZIkiIyPl5uamzp0764UXXrD7DLKCnjuF0b17d+3du1erVq3SwoULVaNGDYWGhqpLly5au3atFixYoPPnz6tatWrq1atXng+0tpqHh4c+/PBDzZo1S19++aU++OADVaxYUV5eXubnhe3Zs0dz5sxRz549dd9995nrDhgwQBs3btTYsWO1bt06VahQQQ8//LB69eqldevW6ZNPPpFhGObnEK5YsUJLly5VRkaGqlSpokceeURDhw4t8CRBAG4tDsbt8GQ1AAAAANyC+FULAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABbhc8iKSG5urk6ePKly5crJwcHB6nYAAAAAWMQwDJ07d05Vq1b9288QJJAVkZMnTyokJMTqNgAAAADcJLZs2aJq1apds4ZAVkTKlSsn6Y833dXV1eJuAAAAAFglMzNTISEhZka4FgJZEblym6KrqyuBDAAAAECBHmViUg8AAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLlLK6ARSP0+ezdD4rJ9+xsi5OqljW5QZ3BAAAAODPCGS3qfNZOXphRWK+Y9M6B6hi2RvbDwAAAIC8uGURAAAAACxCIAMAAAAAixDIAAAAAMAilgaynJwcTZ8+XaGhofLz81PLli01a9YsGYZh1hiGobfffls2m01+fn7q27evfv75Z7vtnD59WsOHD1fjxo3VtGlTjRkzRufOnbOr2bdvn7p3765GjRopJCREc+fOzdPPhg0b1KZNGzVq1Ejt27fXli1biuW4AQAAAECyOJDNnTtXH3zwgV5++WWtX79eI0aM0Lx58xQdHW1XEx0drXHjxmnFihUqU6aM+vfvr0uXLpk1I0aM0KFDh7RgwQLNnj1bO3fu1Msvv2yOZ2Zmqn///qpevbpWrVqlkSNHaubMmVq+fLlZ8/3332v48OF64okntGbNGj300EMaPHiwDhw4cGPeDAAAAAAljqWBLCEhQQ899JAeeOAB1axZU23atJHNZlNSUpKkP66OLV68WM8++6xatmwpHx8fTZ06VSdPntRXX30lSTp8+LC2bt2qiRMnyt/fX02bNtXYsWO1bt06nThxQpL0ySefKDs7W5MmTdI999yjdu3aqVevXlqwYIHZy+LFi9W8eXM99dRT8vT01PPPP68GDRpoyZIlN/6NAQAAAFAiWBrIAgMDFR8fryNHjkj647bCXbt2qUWLFpKk48ePKzU1VcHBweY65cuXl7+/vxISEiT9EeoqVKigRo0amTXBwcFydHQ0g11iYqKaNm0qF5f/++wtm82mI0eO6MyZM2ZNUFCQXX82m02JiYlFf+AAAAAAIIs/h2zAgAHKzMzUI488IicnJ+Xk5Og///mPOnToIElKTU2VJLm7u9ut5+7urrS0NElSWlqaKleubDdeqlQpubm5meunpaWpZs2adjUeHh7mmJubm9LS0sxl+e0HAAAAAIqapYFsw4YNWrt2rd588015eXlp7969ioyMVNWqVRUWFmZlawAAAABQ7CwNZFOnTtWAAQPUrl07SZK3t7dSUlI0Z84chYWFqUqVKpKk9PR0Va1a1VwvPT1dPj4+kv640nXq1Cm77V6+fFlnzpwx1/fw8MhzpevK6ytXxfKrSU9Pz3PVDAAAAACKiqXPkF28eFEODg52y5ycnMxp72vWrKkqVaooLi7OHM/MzNQPP/ygwMBASX88h5aRkaHdu3ebNfHx8crNzZWfn58kKSAgQDt37lR2drZZExsbqzp16sjNzc2siY+Pt+slNjZWAQEBRXfAAAAAAHAVSwPZgw8+qNmzZ2vz5s06fvy4vvzySy1YsEAtW7aUJDk4OKh379569913tXHjRu3fv18jR45U1apVzRpPT081b95c//3vf5WUlKRdu3bp1VdfVbt27XTnnXdKktq3by9nZ2e99NJLOnjwoNavX6/FixerX79+Zi+9e/fW1q1b9f777+vw4cOaMWOGdu/erZ49e974NwYAAABAieBgXP0pzDdYZmam3n77bX311VfmbYnt2rXT4MGDzRkRDcNQVFSUVqxYoYyMDDVp0kSvvPKK6tSpY27n9OnTevXVV/X111/L0dFRDz/8sMaOHaty5cqZNfv27dOECRP0448/qlKlSurZs6cGDBhg18+GDRs0ffp0/frrr6pdu7ZefPFFhYSEFPhYmjRpol27dsnV1bUI3p1/JuX0Bb2wIjHfsWmdA1S9Ypkb2xAAAABQQhQmG1gayG4nBDIAAAAAUuGygaW3LAIAAABASUYgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAilgay0NBQeXt75/kaP368JOnSpUsaP368mjVrpsDAQA0dOlRpaWl220hJSdGAAQPk7++voKAgTZkyRZcvX7ar2bFjh8LCwuTr66tWrVpp1apVeXqJiYlRaGioGjVqpE6dOikpKan4DhwAAAAAZHEg++ijj7Rt2zbza8GCBZKkNm3aSJImTZqkTZs2afr06YqOjtbJkyc1ZMgQc/2cnBwNHDhQ2dnZWrZsmSZPnqzVq1crKirKrDl27JgGDhyoZs2a6eOPP1afPn00duxYbd261axZv369IiMjNXjwYK1evVo+Pj7q37+/0tPTb9A7AQAAAKAksjSQVa5cWVWqVDG/Nm3apLvuukv//ve/dfbsWa1cuVIREREKCgqSr6+vJk2apISEBCUmJkqStm3bpkOHDun1119X/fr1FRISoueee04xMTHKysqSJC1btkw1a9ZURESEPD091bNnT7Vu3VoLFy40+1iwYIE6d+6s8PBweXl5afz48SpdurRWrlxpwbsCAAAAoKS4aZ4hy8rK0ieffKLw8HA5ODho9+7dys7OVnBwsFnj6emp6tWrm4EsMTFR9erVk4eHh1ljs9mUmZmpQ4cOmTVBQUF2+7LZbOY2srKytGfPHrv9ODo6Kjg4WAkJCcV0tAAAAABwEwWyr776SmfPnlVYWJgkKS0tTc7OzqpQoYJdnbu7u1JTU82aq8OYJPP139VkZmbq4sWL+v3335WTkyN3d/c8+/nz82oAAAAAUJRumkC2cuVKtWjRQnfeeafVrQAAAADADXFTBLJff/1VsbGxeuKJJ8xlHh4eys7OVkZGhl1tenq6qlSpYtb8+SrWldd/V+Pq6qrSpUurUqVKcnJyyjOBR3p6ep4rawAAAABQlG6KQLZq1Sq5u7vrgQceMJf5+vrK2dlZcXFx5rLk5GSlpKQoICBAkhQQEKADBw7YhanY2Fi5urrKy8vLrImPj7fbX2xsrLkNFxcXNWzY0G4/ubm5iouLU2BgYBEfKQAAAAD8H8sDWW5urlatWqXHH39cpUqVMpeXL19e4eHhmjx5suLj47V7926NGTNGgYGBZpiy2Wzy8vLSyJEjtW/fPm3dulXTp09Xjx495OLiIknq2rWrjh07pqlTp+rw4cOKiYnRhg0b1LdvX3Nf/fr104oVK7R69WodPnxY48aN04ULF9SxY8cb+VYAAAAAKGFK/X1J8YqNjVVKSorCw8PzjI0ZM0aOjo4aNmyYsrKyZLPZ9Morr5jjTk5Omj17tsaNG6cuXbqoTJkyCgsL07Bhw8yaWrVqac6cOYqMjNTixYtVrVo1TZw4Uc2bNzdr2rZtq1OnTikqKkqpqamqX7++5s2bxy2LAAAAAIqVg2EYhtVN3A4yMzPVpEkT7dq1S66urla3o5TTF/TCisR8x6Z1DlD1imVubEMAAABACVGYbGD5LYsAAAAAUFIRyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCKWB7ITJ05oxIgRatasmfz8/NS+fXv9+OOP5rhhGHr77bdls9nk5+envn376ueff7bbxunTpzV8+HA1btxYTZs21ZgxY3Tu3Dm7mn379ql79+5q1KiRQkJCNHfu3Dy9bNiwQW3atFGjRo3Uvn17bdmypViOGQAAAAAkiwPZmTNn1K1bNzk7O2vu3Llat26dRo0aJTc3N7Nm7ty5io6O1rhx47RixQqVKVNG/fv316VLl8yaESNG6NChQ1qwYIFmz56tnTt36uWXXzbHMzMz1b9/f1WvXl2rVq3SyJEjNXPmTC1fvtys+f777zV8+HA98cQTWrNmjR566CENHjxYBw4cuDFvBgAAAIASx9JANnfuXFWrVk2RkZHy8/NTrVq1ZLPZdNddd0n64+rY4sWL9eyzz6ply5by8fHR1KlTdfLkSX311VeSpMOHD2vr1q2aOHGi/P391bRpU40dO1br1q3TiRMnJEmffPKJsrOzNWnSJN1zzz1q166devXqpQULFpi9LF68WM2bN9dTTz0lT09PPf/882rQoIGWLFly498YAAAAACWCpYHs66+/lq+vr4YNG6agoCA9/vjjWrFihTl+/PhxpaamKjg42FxWvnx5+fv7KyEhQZKUkJCgChUqqFGjRmZNcHCwHB0dlZSUJElKTExU06ZN5eLiYtbYbDYdOXJEZ86cMWuCgoLs+rPZbEpMTCzy4wYAAAAAyeJAduzYMX3wwQeqXbu25s+fr27dumnixIlavXq1JCk1NVWS5O7ubreeu7u70tLSJElpaWmqXLmy3XipUqXk5uZmrp+WliYPDw+7miuvr97On2uu3g8AAAAAFLVSVu7cMAz5+vrqhRdekCQ1aNBABw8e1LJlyxQWFmZlawAAAABQ7Cy9QlalShV5enraLatbt65SUlLMcUlKT0+3q0lPTzevZnl4eOjUqVN245cvX9aZM2fM9T08PPJc6bry+urt/Lnm6v0AAAAAQFGzNJA1btxYR44csVv2888/q0aNGpKkmjVrqkqVKoqLizPHMzMz9cMPPygwMFCSFBgYqIyMDO3evdusiY+PV25urvz8/CRJAQEB2rlzp7Kzs82a2NhY1alTx5zRMSAgQPHx8Xa9xMbGKiAgoOgOGAAAAACuYmkg69Onj3744QfNnj1bR48e1dq1a7VixQp1795dkuTg4KDevXvr3Xff1caNG7V//36NHDlSVatWVcuWLSVJnp6eat68uf773/8qKSlJu3bt0quvvqp27drpzjvvlCS1b99ezs7Oeumll3Tw4EGtX79eixcvVr9+/cxeevfura1bt+r999/X4cOHNWPGDO3evVs9e/a88W8MAAAAgBLBwTAMw8oGNm3apGnTpunnn39WzZo11a9fP3Xu3NkcNwxDUVFRWrFihTIyMtSkSRO98sorqlOnjllz+vRpvfrqq/r666/l6Oiohx9+WGPHjlW5cuXMmn379mnChAn68ccfValSJfXs2VMDBgyw62XDhg2aPn26fv31V9WuXVsvvviiQkJCCnQcmZmZatKkiXbt2iVXV9d/+K78cymnL+iFFYn5jk3rHKDqFcvc2IYAAACAEqIw2cDyQHa7IJABAAAAkAqXDSy9ZREAAAAASjICGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWKVXYFf73v//JwcFB1apVkyQlJSVp7dq18vLyUpcuXYq8QQAAAAC4XRX6Ctnw4cMVHx8vSUpNTVW/fv30448/6q233tLMmTOLvEEAAAAAuF0VOpAdPHhQfn5+kqQNGzbonnvu0bJly/TGG29o9erVRd4gAAAAANyuCh3ILl++LBcXF0lSbGysQkNDJUl169ZVampq0XYHAAAAALexQgcyLy8vLVu2TDt37lRsbKxatGghSTp58qQqVqxY1P0BAAAAwG2r0IFsxIgRWr58uXr16qV27drJx8dHkvT111+btzICAAAAAP5eoWdZbNasmeLj45WZmSk3NzdzeefOnVW2bNkibQ4AAAAAbmeFvkLWu3dvnTt3zi6MSVLFihX1/PPPF1VfAAAAAHDbK3Qg+/bbb5WdnZ1n+aVLl7Rr164iaQoAAAAASoIC37K4b98+88+HDh2ym1ExNzdXW7du1Z133lm03QEAAADAbazAgezxxx+Xg4ODHBwc1KdPnzzjpUuX1tixY4u0OQAAAAC4nRU4kG3cuFGGYahly5b68MMPVblyZXPM2dlZ7u7ucnJyKpYmAQAAAOB2VOBnyGrUqKGaNWtq3759atSokWrUqGF+Va1aVU5OTjIMo1A7nzFjhry9ve2+2rRpY45funRJ48ePV7NmzRQYGKihQ4cqLS3NbhspKSkaMGCA/P39FRQUpClTpujy5ct2NTt27FBYWJh8fX3VqlUrrVq1Kk8vMTExCg0NVaNGjdSpUyclJSUV6lgAAAAAoLAKPalHRESEzp8/n2f58ePH1aNHj0I3cM8992jbtm3m19KlS82xSZMmadOmTZo+fbqio6N18uRJDRkyxBzPycnRwIEDlZ2drWXLlmny5MlavXq1oqKizJpjx45p4MCBatasmT7++GP16dNHY8eO1datW82a9evXKzIyUoMHD9bq1avl4+Oj/v37Kz09vdDHAwAAAAAFVehAtm/fPnXo0EEJCQnmstWrV+uxxx5TpUqVCt2Ak5OTqlSpYn5duRXy7NmzWrlypSIiIhQUFCRfX19NmjRJCQkJSkxMlCRt27ZNhw4d0uuvv6769esrJCREzz33nGJiYpSVlSVJWrZsmWrWrKmIiAh5enqqZ8+eat26tRYuXGj2sGDBAnXu3Fnh4eHy8vLS+PHjVbp0aa1cubLQxwMAAAAABVXoQPbhhx+qVatW6tWrl6ZNm6bnnntOEyZM0KhRozRr1qxCN3D06FHZbDY99NBDGj58uFJSUiRJu3fvVnZ2toKDg81aT09PVa9e3QxkiYmJqlevnjw8PMwam82mzMxMHTp0yKwJCgqy26fNZjO3kZWVpT179tjtx9HRUcHBwXahEwAAAACKWoEn9bjC2dlZo0aNUpkyZfTOO++oVKlSio6OVmBgYKF37ufnp8jISNWpU0epqamaNWuWevToobVr1yotLU3Ozs6qUKGC3Tru7u7mlPtpaWl2YUyS+frvajIzM3Xx4kWdOXNGOTk5cnd3z7Of5OTkQh8TAAAAABRUoQNZdna23nzzTcXExGjgwIHatWuXhg4dqtdee00hISGF2tbV9T4+PvL399eDDz6oDRs2qHTp0oVtDQAAAABuKYUOZOHh4bp48aKio6MVEBAgwzA0d+5cDRkyROHh4Ro3btx1N1OhQgXVrl1bv/zyi4KDg5Wdna2MjAy7q2Tp6emqUqWKpD+udP15NsQrszBeXfPnmRnT0tLk6uqq0qVLy9HRUU5OTnkm8EhPT89zZQ0AAAAAilKhnyHz9fXVmjVrFBAQIElycHDQgAEDtHz5cu3cufMfNXPu3DkdO3ZMVapUka+vr5ydnRUXF2eOJycnKyUlxdx3QECADhw4YBemYmNj5erqKi8vL7MmPj7ebj+xsbHmNlxcXNSwYUO7/eTm5iouLu66bsMEAAAAgIIq9BWySZMm5bu8QYMG+X6+17VMmTJFDz74oKpXr66TJ09qxowZcnR01KOPPqry5csrPDxckydPlpubm1xdXTVx4kQFBgaaYcpms8nLy0sjR47Uiy++qNTUVE2fPl09evSQi4uLJKlr166KiYnR1KlTFR4ervj4eG3YsEFz5swx++jXr59GjRolX19f+fn5adGiRbpw4YI6duxY2LcHAAAAAAqs0IFMktasWaNly5bp+PHjWr58uWrUqKGFCxeqZs2aatmyZYG389tvv+mFF17Q6dOnVblyZTVp0kQrVqwwp74fM2aMHB0dNWzYMGVlZclms+mVV14x13dyctLs2bM1btw4denSRWXKlFFYWJiGDRtm1tSqVUtz5sxRZGSkFi9erGrVqmnixIlq3ry5WdO2bVudOnVKUVFRSk1NVf369TVv3jxuWQQAAABQrBwMwzAKs8LSpUsVFRWlPn36aPbs2fr0009Vq1YtrVq1SqtXr1Z0dHRx9XpTy8zMVJMmTbRr1y65urpa3Y5STl/QCysS8x2b1jlA1SuWubENAQAAACVEYbJBoZ8hW7JkiSZOnKhnn31Wjo7/t7qvr68OHDhQ+G4BAAAAoIQqdCA7fvy46tevn2e5i4uLLly4UCRNAQAAAEBJUOhAVrNmTe3duzfP8q1bt8rT07NImgIAAACAkqDAgWzmzJm6cOGC+vXrpwkTJmj9+vWSpKSkJL377ruaNm2annrqqWJrFAAAAABuNwWeZXHWrFnq1q2bOnXqpDvuuEPTp0/XhQsXNHz4cFWtWlVjxoxRu3btirNXAAAAALitFDiQXT0ZY4cOHdShQwdduHBB58+fl7u7e7E0BwAAAAC3s0J9DpmDg4Pd6zJlyqhMGaZPBwAAAIDrUahA1rp16zyh7M++/fbbf9QQAAAAAJQUhQpkQ4cOVfny5YurFwAAAAAoUQoVyNq1a8fzYgAAAABQRAo87f3f3aoIAAAAACicAgeyq2dZBAAAAAD8cwW+ZXHfvn3F2QcAAAAAlDgFvkIGAAAAAChaBDIAAAAAsAiBDAAAAAAsUqBAFhYWpjNnzkiSZs6cqQsXLhRrUwAAAABQEhQokB0+fNgMYbNmzdL58+eLtSkAAAAAKAkKNMti/fr1NXr0aDVp0kSGYWj+/PkqW7ZsvrVDhgwp0gYBAAAA4HZVoEAWGRmpGTNmaNOmTXJwcNDWrVvl5OSUp87BwYFABgAAAAAFVKBAVrduXb311luSJB8fHy1cuFDu7u7F2hgAAAAA3O4K/MHQV/AB0QAAAABQNAodyCTpl19+0aJFi3T48GFJkpeXl3r37q277rqrSJsDAAAAgNtZoT+HbOvWrWrbtq2SkpLk7e0tb29v/fDDD2rXrp22b99eHD0CAAAAwG2p0FfI3nzzTfXt21cjRoywW/7GG2/ojTfe0P33319kzQEAAADA7azQV8gOHz6sJ554Is/y8PBwHTp0qEiaAgAAAICSoNCBrHLlytq7d2+e5Xv37mXmRQAAAAAohELfstipUye9/PLLOnbsmBo3bixJ+v777zV37lz17du3qPsDAAAAgNtWoQPZ4MGD5erqqvfff1/Tpk2TJFWtWlVDhgxR7969i7xBAAAAALhdFTqQOTg4qG/fvurbt68yMzMlSa6urkXeGAAAAADc7q7rc8iuIIgBAAAAwPUr9KQeAAAAAICiQSADAAAAAIsQyAAAAADAIoUKZNnZ2erTp49+/vnnYmoHAAAAAEqOQgUyZ2dn7d+/v7h6AQAAAIASpdC3LHbo0EEfffRRcfQCAAAAACVKoae9z8nJ0QcffKDY2Fj5+vqqTJkyduOjR48usuYAAAAA4HZW6EB24MABNWjQQJJ05MgRuzEHB4ei6QoAAAAASoBCB7Lo6Oji6AMAAAAASpzrnvb+6NGj2rp1qy5evChJMgyjyJoCAAAAgJKg0FfIfv/9dz3//PPasWOHHBwc9MUXX6hWrVoaM2aM3NzcFBERURx9AgAAAMBtp9BXyCIjI1WqVClt3rxZpUuXNpe3bdtWW7duLdLmAAAAAOB2VuhAtn37dr344ouqVq2a3fLatWsrJSXluht577335O3trddee81cdunSJY0fP17NmjVTYGCghg4dqrS0NLv1UlJSNGDAAPn7+ysoKEhTpkzR5cuX7Wp27NihsLAw+fr6qlWrVlq1alWe/cfExCg0NFSNGjVSp06dlJSUdN3HAgAAAAAFUehAdv78ebsrY1ecPn1aLi4u19VEUlKSli1bJm9vb7vlkyZN0qZNmzR9+nRFR0fr5MmTGjJkiDmek5OjgQMHKjs7W8uWLdPkyZO1evVqRUVFmTXHjh3TwIED1axZM3388cfq06ePxo4da3c1b/369YqMjNTgwYO1evVq+fj4qH///kpPT7+u4wEAAACAgih0IGvatKnWrFljtyw3N1fz5s1Ts2bNCt3AuXPn9OKLL2rixIlyc3Mzl589e1YrV65URESEgoKC5Ovrq0mTJikhIUGJiYmSpG3btunQoUN6/fXXVb9+fYWEhOi5555TTEyMsrKyJEnLli1TzZo1FRERIU9PT/Xs2VOtW7fWwoULzX0tWLBAnTt3Vnh4uLy8vDR+/HiVLl1aK1euLPTxAAAAAEBBFTqQvfjii1qxYoWeeuopZWdn6/XXX9ejjz6qnTt3asSIEYVuYMKECQoJCVFwcLDd8t27dys7O9tuuaenp6pXr24GssTERNWrV08eHh5mjc1mU2Zmpg4dOmTWBAUF2W3bZrOZ28jKytKePXvs9uPo6Kjg4GAlJCQU+ngAAAAAoKAKPctivXr19Pnnn2vJkiUqV66czp8/r1atWqlHjx6qWrVqoba1bt06/fTTT/roo4/yjKWlpcnZ2VkVKlSwW+7u7q7U1FSz5uowJsl8/Xc1mZmZunjxos6cOaOcnBy5u7vn2U9ycnKhjgcAAAAACqPQgUySypcvr2efffYf7fh///ufXnvtNb3//vu64447/tG2AAAAAOBWdF2B7MyZM/roo490+PBhSZKXl5c6duyoihUrFngbe/bsUXp6ujp27Gguy8nJ0XfffaeYmBjNnz9f2dnZysjIsLtKlp6eripVqkj640rXn2dDvDIL49U1f56ZMS0tTa6uripdurQcHR3l5OSUZwKP9PT0PFfWAAAAAKAoFfoZsu+++06hoaGKjo5WRkaGMjIyFB0drYceekjfffddgbdz3333ae3atVqzZo355evrq/bt25t/dnZ2VlxcnLlOcnKyUlJSFBAQIEkKCAjQgQMH7MJUbGysXF1d5eXlZdbEx8fb7Ts2NtbchouLixo2bGi3n9zcXMXFxSkwMLCwbw8AAAAAFFihr5BNmDBBbdu21bhx4+Tk5CTpjytb48eP14QJE7R27doCbcfV1VX16tWzW1a2bFlVrFjRXB4eHq7JkyfLzc1Nrq6umjhxogIDA80wZbPZ5OXlpZEjR+rFF19Uamqqpk+frh49ephT8Hft2lUxMTGaOnWqwsPDFR8frw0bNmjOnDnmfvv166dRo0bJ19dXfn5+WrRokS5cuGB39Q4AAAAAilqhA9nRo0f19ttvm2FMkpycnNS3b9880+H/U2PGjJGjo6OGDRumrKws2Ww2vfLKK3b7nT17tsaNG6cuXbqoTJkyCgsL07Bhw8yaWrVqac6cOYqMjNTixYtVrVo1TZw4Uc2bNzdr2rZtq1OnTikqKkqpqamqX7++5s2bxy2LAAAAAIqVg2EYRmFW6Nq1q5566im1bNnSbvlXX32l9957TytWrCjSBm8VmZmZatKkiXbt2iVXV1er21HK6Qt6YUVivmPTOgeoesUyN7YhAAAAoIQoTDYo0BWyffv2mX/u3bu3XnvtNR09elT+/v6SpB9++EExMTHX9TlkAAAAAFBSFSiQPf7443JwcNDVF9Nef/31PHXDhw9X27Zti647AAAAALiNFSiQbdy4sbj7AAAAAIASp0CBrEaNGsXdBwAAAACUONf1wdAnTpzQrl27dOrUKeXm5tqN9e7du0gaAwAAAIDbXaED2apVq/Tyyy/L2dlZlSpVshtzcHAgkAEAAABAARU6kL399tsaPHiwBg4cKEdHx+LoCQAAAABKhEIHsosXL6pdu3aEsVuYk6ODUk5fyHesrIuTKpZ1ucEdAQAAACVToQNZeHi4PvvsMw0YMKA4+sENcCk7RyNXJuU7Nq1zgCqWvcENAQAAACVUoQPZ8OHDNXDgQG3dulX16tVTqVL2mxg9enSRNQcAAAAAt7NCB7I5c+Zo27ZtqlOnTp4xBweHImkKAAAAAEqCQgeyBQsWaNKkSerYsWNx9AMAAAAAJUahZ+ZwcXFR48aNi6MXAAAAAChRCh3IevfurSVLlhRHLwAAAABQohT6lsWkpCTFx8dr06ZNuueee/JM6jFz5swiaw4AAAAAbmeFDmQVKlTQww8/XBy9AAAAAECJUuhAFhkZWRx9AAAAAECJU+hnyAAAAAAARaPQV8hCQ0Ov+XljGzdu/EcNAQAAAEBJUehA1qdPH7vXly9f1k8//aRt27apf//+RdYYAAAAANzu/nEguyImJka7d+/+xw0BAAAAQElRZM+QtWjRQp9//nlRbQ4AAAAAbntFFsg+++wzVaxYsag2BwAAAAC3vULfsvj444/bTephGIbS0tJ06tQpvfLKK0XaHAAAAADczgodyFq2bGn32sHBQZUrV9a///1veXp6FlljAAAAAHC7K3QgGzJkSHH0AQAAAAAlDh8MDQAAAAAWKfAVMh8fn2t+ILT0x+2LP/300z9uCgAAAABKggIHspkzZ/7lWGJioqKjo5Wbm1skTQEAAABASVDgQPbnyTwkKTk5WW+++aY2bdqk9u3ba9iwYUXaHAAAAADczgo9qYcknThxQjNmzNCaNWtks9m0Zs0a1atXr6h7AwAAAIDbWqEC2dmzZzV79mwtWbJE9evX18KFC9W0adPi6g0AAAAAbmsFDmRz587VvHnz5OHhoTfffDPfWxgBAAAAAAVX4ED25ptvqnTp0rrrrru0Zs0arVmzJt+6a03+AQAAAAD4PwUOZI8//vjfTnsPAAAAACi4AgeyyZMnF2cfAAAAAFDiOFrdAAAAAACUVAQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxiaSBbunSp2rdvr8aNG6tx48bq0qWLtmzZYo5funRJ48ePV7NmzRQYGKihQ4cqLS3NbhspKSkaMGCA/P39FRQUpClTpujy5ct2NTt27FBYWJh8fX3VqlUrrVq1Kk8vMTExCg0NVaNGjdSpUyclJSUVz0EDAAAAwP9naSCrVq2aRowYoVWrVmnlypW67777NHjwYB08eFCSNGnSJG3atEnTp09XdHS0Tp48qSFDhpjr5+TkaODAgcrOztayZcs0efJkrV69WlFRUWbNsWPHNHDgQDVr1kwff/yx+vTpo7Fjx2rr1q1mzfr16xUZGanBgwdr9erV8vHxUf/+/ZWenn7j3gwAAAAAJY6lgSw0NFQhISGqXbu26tSpo//85z8qW7asEhMTdfbsWa1cuVIREREKCgqSr6+vJk2apISEBCUmJkqStm3bpkOHDun1119X/fr1FRISoueee04xMTHKysqSJC1btkw1a9ZURESEPD091bNnT7Vu3VoLFy40+1iwYIE6d+6s8PBweXl5afz48SpdurRWrlxpwbsCAAAAoKS4aZ4hy8nJ0bp163T+/HkFBgZq9+7dys7OVnBwsFnj6emp6tWrm4EsMTFR9erVk4eHh1ljs9mUmZmpQ4cOmTVBQUF2+7LZbOY2srKytGfPHrv9ODo6Kjg4WAkJCcV0tAAAAAAglbK6gf3796tr1666dOmSypYtq1mzZsnLy0t79+6Vs7OzKlSoYFfv7u6u1NRUSVJaWppdGJNkvv67mszMTF28eFFnzpxRTk6O3N3d8+wnOTm5SI8VAAAAAK5meSCrU6eO1qxZo7Nnz+rzzz/XqFGjtGTJEqvbAgAAAIBiZ3kgc3Fx0d133y1J8vX11Y8//qjFixfrkUceUXZ2tjIyMuyukqWnp6tKlSqS/rjS9efZEK/Mwnh1zZ9nZkxLS5Orq6tKly4tR0dHOTk55ZnAIz09Pc+VNQAAAAAoSjfNM2RX5ObmKisrS76+vnJ2dlZcXJw5lpycrJSUFAUEBEiSAgICdODAAbswFRsbK1dXV3l5eZk18fHxdvuIjY01t+Hi4qKGDRva7Sc3N1dxcXEKDAwspqMEAAAAAIuvkL355ptq0aKF/vWvf+ncuXP69NNP9e2332r+/PkqX768wsPDNXnyZLm5ucnV1VUTJ05UYGCgGaZsNpu8vLw0cuRIvfjii0pNTdX06dPVo0cPubi4SJK6du2qmJgYTZ06VeHh4YqPj9eGDRs0Z84cs49+/fpp1KhR8vX1lZ+fnxYtWqQLFy6oY8eOVrwtAAAAAEoISwNZenq6Ro0apZMnT6p8+fLy9vbW/Pnzdf/990uSxowZI0dHRw0bNkxZWVmy2Wx65ZVXzPWdnJw0e/ZsjRs3Tl26dFGZMmUUFhamYcOGmTW1atXSnDlzFBkZqcWLF6tatWqaOHGimjdvbta0bdtWp06dUlRUlFJTU1W/fn3NmzePWxYBAAAAFCsHwzAMq5u4HWRmZqpJkybatWuXXF1drW5HKacv6IUVifmOTQ3308iVSfmOTescoOoVyxRjZwAAAMDtrTDZ4KZ7hgwAAAAASgoCGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGARAhkAAAAAWIRABgAAAAAWsTSQzZkzR+Hh4QoMDFRQUJAGDRqk5ORku5pLly5p/PjxatasmQIDAzV06FClpaXZ1aSkpGjAgAHy9/dXUFCQpkyZosuXL9vV7NixQ2FhYfL19VWrVq20atWqPP3ExMQoNDRUjRo1UqdOnZSUlFT0Bw0AAAAA/5+lgezbb79Vjx49tGLFCi1YsECXL19W//79df78ebNm0qRJ2rRpk6ZPn67o6GidPHlSQ4YMMcdzcnI0cOBAZWdna9myZZo8ebJWr16tqKgos+bYsWMaOHCgmjVrpo8//lh9+vTR2LFjtXXrVrNm/fr1ioyM1ODBg7V69Wr5+Piof//+Sk9PvzFvBgAAAIASx9JANn/+fHXs2FH33HOPfHx8NHnyZKWkpGjPnj2SpLNnz2rlypWKiIhQUFCQfH19NWnSJCUkJCgxMVGStG3bNh06dEivv/666tevr5CQED333HOKiYlRVlaWJGnZsmWqWbOmIiIi5OnpqZ49e6p169ZauHCh2cuCBQvUuXNnhYeHy8vLS+PHj1fp0qW1cuXKG/22AAAAACghbqpnyM6ePStJcnNzkyTt3r1b2dnZCg4ONms8PT1VvXp1M5AlJiaqXr168vDwMGtsNpsyMzN16NAhsyYoKMhuXzabzdxGVlaW9uzZY7cfR0dHBQcHKyEhociPEwAAAACkmyiQ5ebmatKkSWrcuLHq1asnSUpLS5Ozs7MqVKhgV+vu7q7U1FSz5uowJsl8/Xc1mZmZunjxon7//Xfl5OTI3d09z37+/LwaAAAAABSVUlY3cMX48eN18OBBLV261OpWAAAAAOCGuCmukE2YMEGbN2/WokWLVK1aNXO5h4eHsrOzlZGRYVefnp6uKlWqmDV/vop15fXf1bi6uqp06dKqVKmSnJyc8kzgkZ6enufKGgAAAAAUFUsDmWEYmjBhgr788kstWrRItWrVshv39fWVs7Oz4uLizGXJyclKSUlRQECAJCkgIEAHDhywC1OxsbFydXWVl5eXWRMfH2+37djYWHMbLi4uatiwod1+cnNzFRcXp8DAwKI8ZAAAAAAwWRrIxo8fr08++URvvvmmypUrp9TUVKWmpurixYuSpPLlyys8PFyTJ09WfHy8du/erTFjxigwMNAMUzabTV5eXho5cqT27dunrVu3avr06erRo4dcXFwkSV27dtWxY8c0depUHT58WDExMdqwYYP69u1r9tKvXz+tWLFCq1ev1uHDhzVu3DhduHBBHTt2vNFvCwAAAIASwtJnyD744ANJUq9eveyWR0ZGmkFozJgxcnR01LBhw5SVlSWbzaZXXnnFrHVyctLs2bM1btw4denSRWXKlFFYWJiGDRtm1tSqVUtz5sxRZGSkFi9erGrVqmnixIlq3ry5WdO2bVudOnVKUVFRSk1NVf369TVv3jxuWQQAAABQbBwMwzCsbuJ2kJmZqSZNmmjXrl1ydXW1uh2lnL6gF1Yk5js2NdxPI1cm5Ts2rXOAqlcsU4ydAQAAALe3wmSDm2JSDwAAAAAoiQhkAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYhEAGAAAAABYhkAEAAACARQhkAAAAAGCRUlY3gJuLk6ODUk5f+Mvxsi5OqljW5QZ2BAAAANy+CGSwcyk7RyNXJv3l+LTOAapY9gY2BAAAANzGuGURAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAilgay7777Ts8884xsNpu8vb311Vdf2Y0bhqG3335bNptNfn5+6tu3r37++We7mtOnT2v48OFq3LixmjZtqjFjxujcuXN2Nfv27VP37t3VqFEjhYSEaO7cuXl62bBhg9q0aaNGjRqpffv22rJlS5EfLwAAAABczdJAdv78eXl7e+uVV17Jd3zu3LmKjo7WuHHjtGLFCpUpU0b9+/fXpUuXzJoRI0bo0KFDWrBggWbPnq2dO3fq5ZdfNsczMzPVv39/Va9eXatWrdLIkSM1c+ZMLV++3Kz5/vvvNXz4cD3xxBNas2aNHnroIQ0ePFgHDhwovoMHAAAAUOJZGshCQkL0n//8R61atcozZhiGFi9erGeffVYtW7aUj4+Ppk6dqpMnT5pX0g4fPqytW7dq4sSJ8vf3V9OmTTV27FitW7dOJ06ckCR98sknys7O1qRJk3TPPfeoXbt26tWrlxYsWGDua/HixWrevLmeeuopeXp66vnnn1eDBg20ZMmSG/NGAAAAACiRbtpnyI4fP67U1FQFBweby8qXLy9/f38lJCRIkhISElShQgU1atTIrAkODpajo6OSkpIkSYmJiWratKlcXFzMGpvNpiNHjujMmTNmTVBQkN3+bTabEhMTi+vwAAAAAODmDWSpqamSJHd3d7vl7u7uSktLkySlpaWpcuXKduOlSpWSm5ubuX5aWpo8PDzsaq68vno7f665ej8AAAAAUBxu2kAGAAAAALe7mzaQValSRZKUnp5utzw9Pd28muXh4aFTp07ZjV++fFlnzpwx1/fw8MhzpevK66u38+eaq/cDAAAAAMXhpg1kNWvWVJUqVRQXF2cuy8zM1A8//KDAwEBJUmBgoDIyMrR7926zJj4+Xrm5ufLz85MkBQQEaOfOncrOzjZrYmNjVadOHbm5uZk18fHxdvuPjY1VQEBAcR0eAAAAAFgbyM6dO6e9e/dq7969kv6YyGPv3r1KSUmRg4ODevfurXfffVcbN27U/v37NXLkSFWtWlUtW7aUJHl6eqp58+b673//q6SkJO3atUuvvvqq2rVrpzvvvFOS1L59ezk7O+ull17SwYMHtX79ei1evFj9+vUz++jdu7e2bt2q999/X4cPH9aMGTO0e/du9ezZ88a/KQAAAABKjFJW7nz37t3q3bu3+ToyMlKSFBYWpsmTJ+vpp5/WhQsX9PLLLysjI0NNmjTRvHnzdMcdd5jrvPHGG3r11VfVp08fOTo66uGHH9bYsWPN8fLly2v+/PmaMGGCOnbsqEqVKmnQoEHq0qWLWdO4cWO98cYbmj59uqZNm6batWtr1qxZqlev3g14FwAAAACUVJYGsmbNmmn//v1/Oe7g4KDnnntOzz333F/WVKxYUW+++eY19+Pj46OlS5des+aRRx7RI488cu2GAQAAAKAI3bTPkAEAAADA7Y5ABgAAAAAWIZABAAAAgEUIZAAAAABgEUsn9cCtx8nRQSmnL+Q7VtbFSRXLutzgjgAAAIBbF4EMhXIpO0cjVyblOzatc4Aqlr3BDQEAAAC3MG5ZBAAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLlLK6Adw+nBwdlHL6Qr5jZV2cVLGsyw3uCAAAALi5EchQZC5l52jkyqR8x6Z1DlDFsje4IQAAAOAmxy2LAAAAAGARAhkAAAAAWIRABgAAAAAWIZABAAAAgEUIZAAAAABgEQIZAAAAAFiEQAYAAAAAFiGQAQAAAIBFCGQAAAAAYBECGQAAAABYpJTVDaBkcHJ0UMrpC/mOlXVxUsWyLje4IwAAAMB6BDLcEJeyczRyZVK+Y9M6B6hi2RvcEAAAAHAT4JZFAAAAALAIgQwAAAAALEIgAwAAAACL8AwZLMeEHwAAACipCGSwHBN+AAAAoKTilkUAAAAAsAiBDAAAAAAswi2LuKnxfBkAAABuZwQy3NR4vgwAAAC3MwIZbllcPQMAAMCtjkD2JzExMZo/f75SU1Pl4+Oj//73v/Lz87O6LeTjWlfP3u4a+JdhzcXJUVk5ufmOEeQAAABwIxHIrrJ+/XpFRkZq/Pjx8vf316JFi9S/f3999tlncnd3t7o9FMK1wtrUcD9ugwQAAMBNgUB2lQULFqhz584KDw+XJI0fP16bN2/WypUrNWDAAIu7w41wrdsgr3Vl7WYa4yofAADArYNA9v9lZWVpz549GjhwoLnM0dFRwcHBSkhI+Nv1DcOQJGVmZhZbj4VxLvOCci7lHyzOncu8rrF/su6tMnbq9Bm9/PGefMcmPNbwlhibEu6n1FNGvmM3U3AsQ3AEAAC3qSuZ4EpGuBYHoyBVJcCJEyfUokULLVu2TIGBgebyqVOn6rvvvtOHH354zfV/++03hYSEFHebAAAAAG4RW7ZsUbVq1a5ZwxWyIlK1alVt2bJF5cqVk4ODg9XtAAAAALCIYRg6d+6cqlat+re1BLL/r1KlSnJyclJ6errd8vT0dHl4ePzt+o6Ojn+bfgEAAACUDOXLly9QnWMx93HLcHFxUcOGDRUXF2cuy83NVVxcnN0tjAAAAABQVLhCdpV+/fpp1KhR8vX1lZ+fnxYtWqQLFy6oY8eOVrcGAAAA4DZEILtK27ZtderUKUVFRSk1NVX169fXvHnzCnTLIgAAAAAUFrMsAgAAAIBFeIYMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiB7DYTExOj0NBQNWrUSJ06dVJSUpLVLeEG+u677/TMM8/IZrPJ29tbX331ld24YRh6++23ZbPZ5Ofnp759++rnn3+2qzl9+rSGDx+uxo0bq2nTphozZozOnTtnV7Nv3z51795djRo1UkhIiObOnVvch4ZiNGfOHIWHhyswMFBBQUEaNGiQkpOT7WouXbqk8ePHq1mzZgoMDNTQoUOVlpZmV5OSkqIBAwbI399fQUFBmjJlii5fvmxXs2PHDoWFhcnX11etWrXSqlWriv34UHyWLl2q9u3bq3HjxmrcuLG6dOmiLVu2mOOcNyiI9957T97e3nrttdfMZZw7yM+MGTPk7e1t99WmTRtz/JY9bwzcNtatW2c0bNjQ+Oijj4yDBw8aY8eONZo2bWqkpaVZ3RpukM2bNxvTpk0zvvjiC6NevXrGl19+aTc+Z84co0mTJsaXX35p7N2713jmmWeM0NBQ4+LFi2ZN//79jQ4dOhiJiYnGd999Z7Rq1cp44YUXzPGzZ88awcHBxvDhw40DBw4Yn376qeHn52csW7bshh0nitaTTz5prFy50jhw4ICxd+9e4+mnnzYeeOAB49y5c2bNyy+/bISEhBixsbHGjz/+aHTu3Nno0qWLOX758mXj0UcfNfr27Wv89NNPxubNm41mzZoZb775plnzyy+/GP7+/kZkZKRx6NAhIzo62qhfv77xzTff3NDjRdHZuHGjsXnzZuPIkSNGcnKyMW3aNKNhw4bGgQMHDMPgvMHf++GHH4wHH3zQaN++vTFx4kRzOecO8hMVFWW0a9fOOHnypPmVnp5ujt+q5w2B7DbyxBNPGOPHjzdf5+TkGDabzZgzZ46FXcEqfw5kubm5xv3332/MmzfPXJaRkWH4+voan376qWEYhnHo0CGjXr16RlJSklmzZcsWw9vb2/jtt98MwzCMmJgY49577zUuXbpk1rz++utG69ati/uQcIOkp6cb9erVM7799lvDMP44Txo2bGhs2LDBrLlyriQkJBiG8ccvA3x8fIzU1FSzZunSpUbjxo3Nc2Xq1KlGu3bt7Pb1/PPPG08++WQxHxFupHvvvddYsWIF5w3+VmZmpvHwww8b27dvN3r27GkGMs4d/JWoqCijQ4cO+Y7dyucNtyzeJrKysrRnzx4FBwebyxwdHRUcHKyEhAQLO8PN4vjx40pNTbU7R8qXLy9/f3/zHElISFCFChXUqFEjsyY4OFiOjo7m7a+JiYlq2rSpXFxczBqbzaYjR47ozJkzN+hoUJzOnj0rSXJzc5Mk7d69W9nZ2Xbnjqenp6pXr67ExERJf5wX9erVk4eHh1ljs9mUmZmpQ4cOmTVBQUF2+7LZbOY2cGvLycnRunXrdP78eQUGBnLe4G9NmDBBISEhdueIxL85uLajR4/KZrPpoYce0vDhw5WSkiLp1j5vShXblnFD/f7778rJyZG7u7vdcnd39zzPgqBkSk1NlaR8z5Er91enpaWpcuXKduOlSpWSm5ubuX5aWppq1qxpV3PlH7a0tDTzh3jcmnJzczVp0iQ1btxY9erVk/TH99XZ2VkVKlSwq3V3d7c7L67+D076v/Pi72oyMzN18eJFlS5duliOCcVr//796tq1qy5duqSyZctq1qxZ8vLy0t69ezlv8JfWrVunn376SR999FGeMf7NwV/x8/NTZGSk6tSpo9TUVM2aNUs9evTQ2rVrb+nzhkAGADCNHz9eBw8e1NKlS61uBbeIOnXqaM2aNTp79qw+//xzjRo1SkuWLLG6LdzE/ve//+m1117T+++/rzvuuMPqdnALCQkJMf/s4+Mjf39/Pfjgg9qwYcMtHbC5ZfE2UalSJTk5OSk9Pd1ueXp6ep6Uj5KpSpUqknTNc8TDw0OnTp2yG798+bLOnDljru/h4ZFnxqIrrznXbm0TJkzQ5s2btWjRIlWrVs1c7uHhoezsbGVkZNjVp6enF+i8+LsaV1fXW/o/0pLOxcVFd999t3x9fTV8+HD5+Pho8eLFnDf4S3v27FF6ero6duyoBg0aqEGDBvr2228VHR2tBg0acO6gwCpUqKDatWvrl19+uaXPGwLZbcLFxUUNGzZUXFycuSw3N1dxcXEKDAy0sDPcLGrWrKkqVarYnSOZmZn64YcfzHMkMDBQGRkZ2r17t1kTHx+v3Nxc+fn5SZICAgK0c+dOZWdnmzWxsbGqU6cOtyveogzD0IQJE/Tll19q0aJFqlWrlt24r6+vnJ2d7c6d5ORkpaSkKCAgQNIf58WBAwfsAn9sbKxcXV3l5eVl1sTHx9ttOzY21twGbg+5ubnKysrivMFfuu+++7R27VqtWbPG/PL19VX79u3NP3PuoCDOnTunY8eOqUqVKrf2eVNs04Xghlu3bp3h6+trrFq1yjh06JDx3//+12jatKndTDK4vWVmZho//fST8dNPPxn16tUzFixYYPz000/Gr7/+ahjGH9PeN23a1Pjqq6+Mffv2Gc8++2y+094//vjjxg8//GDs3LnTePjhh+2mvc/IyDCCg4ONF1980Thw4ICxbt06w9/fn2nvb2GvvPKK0aRJE2PHjh12UwlfuHDBrHn55ZeNBx54wIiLizN+/PFHo0uXLvlOJfzkk08ae/fuNb755hvjvvvuy3cq4SlTphiHDh0ylixZwhTUt7g33njD+Pbbb41jx44Z+/btM9544w3D29vb2LZtm2EYnDcouKtnWTQMzh3kb/LkycaOHTuMY8eOGbt27TL69u1rNGvWzJz6/lY9bwhkt5no6GjjgQceMBo2bGg88cQTRmJiotUt4QaKj4836tWrl+dr1KhRhmH8MfX99OnTjeDgYMPX19fo06ePkZycbLeN33//3XjhhReMgIAAo3HjxkZERISRmZlpV7N3716jW7duhq+vr9G8eXM+WuEWl985U69ePWPlypVmzcWLF41x48YZ9957r+Hv728MHjzYOHnypN12jh8/bjz11FOGn5+f0axZM2Py5MlGdna2XU18fLzx2GOPGQ0bNjQeeughu33g1jN69GjjwQcfNBo2bGjcd999Rp8+fcwwZhicNyi4Pwcyzh3k5/nnnzfuv/9+o2HDhkbz5s2N559/3jh69Kg5fqueNw6GYRjFd/0NAAAAAPBXeIYMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsAiBDAAAAAAsQiADAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAcNs7fvy4vL29tXfvXqtbAQDATimrGwAAoCC8vb2vOT5kyBANHTr0BnVTcEePHtXs2bO1fft2nTp1SlWrVlVAQID69eunRo0a3bA+jh8/roceekhr1qxR/fr1b9h+AQDXRiADANwStm3bZv55/fr1ioqK0meffWYuK1u2rBVtXdOPP/6ovn376p577tGECRNUt25dnTt3Ths3btSUKVO0ZMkSq1sEAFiMWxYBALeEKlWqmF/ly5eXg4OD+drd3V0LFixQixYt5Ovrq8cee0zffPPNX24rJydHo0ePVps2bZSSkiJJ+uqrrxQWFqZGjRrpoYce0syZM3X58mVzHW9vb3344YcaPHiw/P399fDDD2vjxo1/uQ/DMDR69GjdfffdWrp0qR544AHdddddql+/voYMGaJ33nnHrN2/f7969+4tPz8/NWvWTP/973917tw5c7xXr1567bXX7LY/aNAgRUREmK9DQ0M1e/ZsjR49WoGBgXrggQe0fPlyc/yhhx6SJD3++OPy9vZWr169/u4tBwDcAAQyAMAtb/HixVqwYIFGjRqlTz75RDabTYMGDdLPP/+cpzYrK0vPPfec9u3bp6VLl6p69erauXOnRo0apd69e2v9+vWaMGGCVq1apdmzZ9utO3PmTD3yyCP65JNP1KJFC40YMUKnT5/Ot6e9e/fq4MGDevLJJ+XomPe/2woVKkiSzp8/r/79+8vNzU0fffSRpk+frtjYWL366quFfh8WLFggX19frVmzRt27d9e4ceOUnJwsSfrwww8lSQsXLtS2bds0Y8aMQm8fAFD0CGQAgFve/Pnz9fTTT6tdu3aqW7euXnzxRfn4+GjRokV2defOndOAAQN06tQpLV68WJUrV5b0R9AaMGCAwsLCVKtWLd1///167rnntGzZMrv1w8LC9Oijj+ruu+/WCy+8oPPnzyspKSnfnq6Ewbp1616z908//VRZWVmaMmWK6tWrp6CgIL388sv6+OOPlZaWVqj3oUWLFurRo4fuvvtuPf3006pUqZJ27NghSeaxVqxYUVWqVFHFihULtW0AQPHgGTIAwC0tMzNTJ0+eVOPGje2WN27cWPv27bNbNnz4cFWrVk2LFi1S6dKlzeX79u3T999/b3dFLCcnR5cuXdKFCxdUpkwZSfYTi5QtW1aurq46derUP+r/8OHD8vb2tnsGrnHjxsrNzdWRI0fk4eFR4G1d3Z+Dg4M8PDyUnp7+j/oDABQvAhkAoMQICQnRJ598ooSEBAUFBZnLz58/r6FDh+rhhx/Os84dd9xh/tnZ2dluzMHBQbm5ufnuq3bt2pKk5ORkNWjQ4B/17eDgIMMw7JZd/XzbFaVK2f+3nt96AICbC7csAgBuaa6urqpataq+//57u+Xff/+9vLy87JZ169ZNw4cP16BBg/Ttt9+ayxs0aKAjR47o7rvvzvOV3/NfBVG/fn15eXnp/fffzze0ZWRkSJI8PT21f/9+nT9/3q53R0dH1alTR9Iftxumpqaa4zk5OTp48GCh+rkSJnNycgp9LACA4kMgAwDc8vr376+5c+dq/fr1Sk5O1htvvKF9+/apd+/eeWp79eql5557TgMHDtTOnTslSYMHD9bHH3+smTNn6uDBgzp8+LDWrVunt95667p7cnBwUGRkpH7++Wd1795dW7Zs0bFjx7Rv3z69++67GjRokCSpffv2cnFxUUREhA4cOKD4+Hi9+uqreuyxx8zbFe+77z5t2bJFmzdv1uHDhzVu3Dgz0BWUu7u7Spcura1btyotLU1nz5697mMDABQdblkEANzyevfurczMTE2ePFmnTp2Sp6en3nnnHfO2wT/r27evDMPQgAEDNG/ePDVv3lyzZ8/WrFmzNHfuXJUqVUp169ZVp06d/lFffn5+WrlypWbPnq2xY8fq999/V9WqVRUYGKgxY8ZIksqUKaP58+frtdde0xNPPKEyZcro4YcftpvSPjw8XPv27dOoUaPk5OSkvn37qlmzZoXqpVSpUho7dqxmzZqlqKgoNW3aVNHR0f/o+AAA/5yDwc3lAAAAAGAJblkEAAAAAIsQyAAAAADAIgQyAAAAALAIgQwAAAAALEIgAwAAAACLEMgAAAAAwCIEMgAAAACwCIEMAAAAACxCIAMAAAAAixDIAAAAAMAiBDIAAAAAsMj/A4wMdR3rAJO4AAAAAElFTkSuQmCC\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "percentiles = [np.percentile(token_counts, p) for p in range(0, 101, 10)]\n",
        "print(percentiles)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sokSkXAmXnkv",
        "outputId": "2d34d55f-6ecc-41b4-bf0a-c038dafd789b"
      },
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[4.0, 15.0, 22.0, 31.0, 40.0, 52.0, 67.0, 89.0, 125.0, 207.0, 4950.0]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "percentile_95 = np.percentile(token_counts, 95)\n",
        "print(percentile_95)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "uHhZn2nFZaVo",
        "outputId": "76f9391b-88b4-4cce-9e06-dffdd239a044"
      },
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "310.5\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "percentiles = [np.percentile(token_counts, p) for p in range(101)]\n",
        "\n",
        "# Plotting\n",
        "plt.figure(figsize=(10, 6))\n",
        "plt.plot(range(101), percentiles)\n",
        "plt.axhline(y=percentile_98, color='r', linestyle='--')\n",
        "plt.axvline(x=95, color='r', linestyle='--')\n",
        "plt.title('Token Count Percentiles')\n",
        "plt.xlabel('Percentile')\n",
        "plt.ylabel('Token Count')\n",
        "plt.grid(True)\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 564
        },
        "id": "Rpav0JyJay3i",
        "outputId": "273ec147-1916-4195-9453-0559e095d58d"
      },
      "execution_count": 43,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1000x600 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **In this model we will use BERT to solve this classification problem**"
      ],
      "metadata": {
        "id": "kQ9H4fSv6PDg"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "data = pd.read_csv(\"/content/train.csv\")"
      ],
      "metadata": {
        "id": "JcgAiLim6r0k"
      },
      "execution_count": 44,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
        "device"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TsLeF_7WP_NC",
        "outputId": "31aa02bc-c1f1-4bf0-d787-1f09ecf315da"
      },
      "execution_count": 45,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "device(type='cuda')"
            ]
          },
          "metadata": {},
          "execution_count": 45
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Initialize the tokenizer\n",
        "tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\n",
        "\n",
        "# Tokenization function\n",
        "def tokenize_data(text, max_length=512):\n",
        "    return tokenizer.encode_plus(\n",
        "        text,\n",
        "        max_length=max_length,\n",
        "        truncation=True,\n",
        "        padding='max_length',\n",
        "        return_tensors='pt'\n",
        "    )\n",
        "\n",
        "# Preprocess the data\n",
        "class CommentDataset(Dataset):\n",
        "    def __init__(self, comments, labels):\n",
        "        self.comments = [tokenize_data(comment) for comment in comments]\n",
        "        self.labels = labels\n",
        "\n",
        "    def __len__(self):\n",
        "        return len(self.comments)\n",
        "\n",
        "    def __getitem__(self, idx):\n",
        "        item = {key: val.squeeze() for key, val in self.comments[idx].items()}\n",
        "        item['labels'] = torch.tensor(self.labels[idx])\n",
        "        return item\n",
        "\n",
        "# Prepare dataset\n",
        "labels = data[['toxic', 'severe_toxic', 'obscene', 'threat', 'insult', 'identity_hate']].values\n",
        "comments = data['comment_text'].tolist()\n",
        "\n",
        "# Split the dataset\n",
        "train_comments, val_comments, train_labels, val_labels = train_test_split(comments, labels, test_size=0.2, random_state=42)\n",
        "\n",
        "# Create datasets\n",
        "train_dataset = CommentDataset(train_comments, train_labels)\n",
        "val_dataset = CommentDataset(val_comments, val_labels)"
      ],
      "metadata": {
        "id": "JKwRMQdD6Vj2"
      },
      "execution_count": 46,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from transformers import BertForSequenceClassification\n",
        "from torch.optim import AdamW\n",
        "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
        "# device = 'cpu'\n",
        "\n",
        "# Load the pre-trained BERT model\n",
        "model = BertForSequenceClassification.from_pretrained(\n",
        "    'bert-base-uncased',\n",
        "    num_labels=6\n",
        ")\n",
        "model.to(device)\n",
        "\n",
        "# Optimizer\n",
        "optimizer = AdamW(model.parameters(), lr=2e-5)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ppvMjxZg8eW8",
        "outputId": "d4c31133-c1b7-4eb8-eac1-60e1901cbf37"
      },
      "execution_count": 47,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Some weights of BertForSequenceClassification were not initialized from the model checkpoint at bert-base-uncased and are newly initialized: ['classifier.weight', 'classifier.bias']\n",
            "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from torch.nn import BCEWithLogitsLoss"
      ],
      "metadata": {
        "id": "7FBcUMNSJdd-"
      },
      "execution_count": 48,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "def train(model, train_loader, optimizer, epochs=1):\n",
        "    model.train()\n",
        "\n",
        "    for epoch in range(epochs):\n",
        "        total_loss = 0\n",
        "        for i, batch in enumerate(train_loader):\n",
        "            optimizer.zero_grad()\n",
        "\n",
        "            input_ids = batch['input_ids'].to(device)\n",
        "            attention_mask = batch['attention_mask'].to(device)\n",
        "            labels = batch['labels'].to(device)\n",
        "\n",
        "            # Forward pass\n",
        "            outputs = model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "\n",
        "            # Compute loss: Ensure labels are of the correct shape\n",
        "            logits = outputs.logits\n",
        "            loss_fn = BCEWithLogitsLoss()\n",
        "            loss = loss_fn(logits, labels.float())\n",
        "\n",
        "            loss.backward()\n",
        "            optimizer.step()\n",
        "            batch_loss = loss.item()\n",
        "            total_loss += loss.item()\n",
        "\n",
        "# Print loss for each batch\n",
        "            print(f'Epoch {epoch+1}, Batch {i+1}/{len(train_loader)}, Batch Loss: {batch_loss}')\n",
        "\n",
        "        # Print total loss for the epoch\n",
        "        print(f'Epoch {epoch+1}, Total Loss: {total_loss / len(train_loader)}')\n",
        "train_loader = DataLoader(train_dataset, batch_size=8, shuffle=True)\n",
        "train(model, train_loader, optimizer)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5TlBOdc0J7S_",
        "outputId": "9acb3862-f1f3-492e-da3e-ea2912424a0b"
      },
      "execution_count": 52,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1, Batch 15936/15957, Batch Loss: 0.03932883217930794\n",
            "Epoch 1, Batch 15937/15957, Batch Loss: 0.006917535327374935\n",
            "Epoch 1, Batch 15938/15957, Batch Loss: 0.0863385945558548\n",
            "Epoch 1, Batch 15939/15957, Batch Loss: 0.0003109783574473113\n",
            "Epoch 1, Batch 15940/15957, Batch Loss: 0.001189853181131184\n",
            "Epoch 1, Batch 15941/15957, Batch Loss: 0.07620935142040253\n",
            "Epoch 1, Batch 15942/15957, Batch Loss: 0.0003191116265952587\n",
            "Epoch 1, Batch 15943/15957, Batch Loss: 0.05924845486879349\n",
            "Epoch 1, Batch 15944/15957, Batch Loss: 0.012598611414432526\n",
            "Epoch 1, Batch 15945/15957, Batch Loss: 0.0009174399892799556\n",
            "Epoch 1, Batch 15946/15957, Batch Loss: 0.03199072554707527\n",
            "Epoch 1, Batch 15947/15957, Batch Loss: 0.04256858676671982\n",
            "Epoch 1, Batch 15948/15957, Batch Loss: 0.0005401548696681857\n",
            "Epoch 1, Batch 15949/15957, Batch Loss: 0.04476992040872574\n",
            "Epoch 1, Batch 15950/15957, Batch Loss: 0.07974638044834137\n",
            "Epoch 1, Batch 15951/15957, Batch Loss: 0.019866935908794403\n",
            "Epoch 1, Batch 15952/15957, Batch Loss: 0.009250661358237267\n",
            "Epoch 1, Batch 15953/15957, Batch Loss: 0.024542227387428284\n",
            "Epoch 1, Batch 15954/15957, Batch Loss: 0.008908836171030998\n",
            "Epoch 1, Batch 15955/15957, Batch Loss: 0.05093453451991081\n",
            "Epoch 1, Batch 15956/15957, Batch Loss: 0.023796387016773224\n",
            "Epoch 1, Batch 15957/15957, Batch Loss: 0.03524591773748398\n",
            "Epoch 1, Total Loss: 0.042526150649653885\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def evaluate(model, val_loader):\n",
        "    model.eval()\n",
        "\n",
        "    total_loss = 0\n",
        "    loss_fn = BCEWithLogitsLoss()\n",
        "\n",
        "    with torch.no_grad():\n",
        "        for batch in val_loader:\n",
        "            input_ids = batch['input_ids'].to(device)\n",
        "            attention_mask = batch['attention_mask'].to(device)\n",
        "            labels = batch['labels'].to(device)\n",
        "            outputs = model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "            logits = outputs.logits\n",
        "            loss = loss_fn(logits, labels.float())\n",
        "            total_loss += loss.item()\n",
        "\n",
        "    print(f'Validation Loss: {total_loss / len(val_loader)}')\n",
        "\n",
        "# Evaluate the model\n",
        "val_loader = DataLoader(val_dataset, batch_size=8)\n",
        "\n",
        "evaluate(model, val_loader)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NzFOxtpaddbF",
        "outputId": "161d157b-dcf3-4af5-c7f6-97e772d773eb"
      },
      "execution_count": 53,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Validation Loss: 0.0392671263384292\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "X_test = pd.read_csv(\"/content/test.csv\")\n",
        "y_test = pd.read_csv(\"/content/test_labels.csv\")"
      ],
      "metadata": {
        "id": "QASsEj4TJ_AZ"
      },
      "execution_count": 54,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "X_test = X_test['comment_text']\n",
        "y_test = y_test[['toxic', 'severe_toxic', 'obscene', 'threat', 'insult', 'identity_hate']]"
      ],
      "metadata": {
        "id": "jAKO7-dXj7k5"
      },
      "execution_count": 55,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "indices_to_remove = y_test[y_test['toxic'] == -1].index\n",
        "\n",
        "y_test_filtered = y_test.drop(indices_to_remove)\n",
        "X_test_filtered = X_test.drop(indices_to_remove)"
      ],
      "metadata": {
        "id": "w8T7_81lkAFE"
      },
      "execution_count": 56,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_comments = X_test_filtered.tolist()\n",
        "\n",
        "# Create a dataset for test data\n",
        "class TestDataset(Dataset):\n",
        "    def __init__(self, comments):\n",
        "        self.comments = [tokenize_data(comment) for comment in comments]\n",
        "\n",
        "    def __len__(self):\n",
        "        return len(self.comments)\n",
        "\n",
        "    def __getitem__(self, idx):\n",
        "        item = {key: val.squeeze().to(device) for key, val in self.comments[idx].items()}\n",
        "        return item\n",
        "\n",
        "test_dataset = TestDataset(test_comments)\n",
        "\n",
        "# Create DataLoader for test data\n",
        "test_loader = DataLoader(test_dataset, batch_size=8)\n"
      ],
      "metadata": {
        "id": "ZXxuwKDjkDux"
      },
      "execution_count": 57,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import accuracy_score\n",
        "\n",
        "def evaluate_model_accuracy(model, data_loader):\n",
        "    model.eval()\n",
        "    all_preds = []\n",
        "    all_labels = []\n",
        "\n",
        "    with torch.no_grad():\n",
        "        for batch in data_loader:\n",
        "            input_ids = batch['input_ids'].to(device)\n",
        "            attention_mask = batch['attention_mask'].to(device)\n",
        "            labels = batch['labels'].to(device)\n",
        "\n",
        "            outputs = model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "            logits = outputs.logits\n",
        "            preds = torch.sigmoid(logits).cpu().numpy() > 0.5  # Converting to binary output\n",
        "            all_preds.extend(preds)\n",
        "            all_labels.extend(labels.cpu().numpy())\n",
        "\n",
        "    # Calculate accuracy\n",
        "    accuracy = accuracy_score(np.array(all_labels), np.array(all_preds))\n",
        "    return accuracy\n",
        "\n",
        "# Assuming y_test_filtered is available and properly formatted\n",
        "y_test_filtered_labels = y_test_filtered.values\n",
        "test_dataset_with_labels = CommentDataset(test_comments, y_test_filtered_labels)\n",
        "test_loader_with_labels = DataLoader(test_dataset_with_labels, batch_size=5)\n",
        "\n",
        "# Get accuracy\n",
        "accuracy = evaluate_model_accuracy(model, test_loader_with_labels)\n",
        "print(f\"Accuracy: {accuracy}\")\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "h1RBrSXCk1QZ",
        "outputId": "da08c5ed-f3f8-40f8-aeee-5bfe9298c16e"
      },
      "execution_count": 58,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 0.8567632623714402\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Custom sentence**\n",
        "We add functionality where we can test our own sentences and the model can give us accurate results in terms of what the content of the message is."
      ],
      "metadata": {
        "id": "CpiZ2_Uv1iJa"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "sen = \"Kanye west is the greatest of all time\"\n",
        "\n"
      ],
      "metadata": {
        "id": "kP2lVbTmlJTE"
      },
      "execution_count": 132,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def preprocess_for_bert(sentence, tokenizer, max_length=512):\n",
        "    \"\"\"Preprocess a single sentence for BERT.\"\"\"\n",
        "    inputs = tokenizer.encode_plus(\n",
        "        sentence,\n",
        "        add_special_tokens=True,\n",
        "        max_length=max_length,\n",
        "        truncation=True,\n",
        "        padding='max_length',\n",
        "        return_tensors='pt'\n",
        "    )\n",
        "    return inputs\n",
        "\n",
        "# Example sentence\n",
        "\n",
        "# Tokenize the sentence\n",
        "inputs = preprocess_for_bert(sen, tokenizer)\n"
      ],
      "metadata": {
        "id": "mZhJlW27o-M8"
      },
      "execution_count": 133,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Move inputs to device\n",
        "input_ids = inputs['input_ids'].to(device)\n",
        "attention_mask = inputs['attention_mask'].to(device)\n"
      ],
      "metadata": {
        "id": "qKamczGWpnVq"
      },
      "execution_count": 134,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model.eval()\n",
        "\n",
        "# Predict\n",
        "with torch.no_grad():\n",
        "    outputs = model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "    logits = outputs.logits\n",
        "    probabilities = torch.sigmoid(logits).squeeze()\n",
        "\n",
        "# Convert probabilities to numpy array\n",
        "probabilities = probabilities.cpu().numpy()"
      ],
      "metadata": {
        "id": "egVeki1rpp8w"
      },
      "execution_count": 135,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "This returns the predicted labels in a list, so just numbers"
      ],
      "metadata": {
        "id": "i2miS88V5UY9"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "threshold = 0.5\n",
        "predicted_labels = (probabilities > threshold).astype(int)\n",
        "\n",
        "# Print predicted labels\n",
        "print(\"Predicted labels:\", predicted_labels)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ABCqeG5-ptz0",
        "outputId": "a4c4a749-8124-4fd7-ea89-cb3ef49fa83b"
      },
      "execution_count": 136,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Predicted labels: [0 0 0 0 0 0]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "labels = ['toxic', 'severe_toxic', 'obscene', 'threat', 'insult', 'identity_hate']\n",
        "\n",
        "predicted_probabilities = list(zip(labels, probabilities))\n",
        "\n",
        "\n",
        "for label, prob in predicted_probabilities:\n",
        "    print(f\"{label}: {prob:.4f}\")\n",
        "threshold = 0.5\n",
        "predicted_labels = (probabilities > threshold).astype(int)\n",
        "# we want the numpy array to match with the names of the class, thus we use a dictionary\n",
        "predicted_labels_dict = dict(zip(labels, predicted_labels))\n",
        "\n",
        "\n",
        "for label, label_class in predicted_labels_dict.items():\n",
        "    print(f\"{label}: {'Yes' if label_class == 1 else 'No'}\")\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "H8VcV4xppxV8",
        "outputId": "e121358a-8d19-4ec6-c1ab-7edf344df166"
      },
      "execution_count": 111,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "toxic: 0.9979\n",
            "severe_toxic: 0.6118\n",
            "obscene: 0.9911\n",
            "threat: 0.0195\n",
            "insult: 0.9603\n",
            "identity_hate: 0.0773\n",
            "toxic: Yes\n",
            "severe_toxic: Yes\n",
            "obscene: Yes\n",
            "threat: No\n",
            "insult: Yes\n",
            "identity_hate: No\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **In this section we look at how we can get some better more detailed statistics\n",
        "\n",
        "\n",
        "\n",
        "**"
      ],
      "metadata": {
        "id": "nAKn1GU-5aky"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def predict(sentence, model, tokenizer, device):\n",
        "    \"\"\"Generate predictions for a single sentence.\"\"\"\n",
        "    inputs = preprocess_for_bert(sentence, tokenizer)\n",
        "    input_ids = inputs['input_ids'].to(device)\n",
        "    attention_mask = inputs['attention_mask'].to(device)\n",
        "\n",
        "    model.eval()\n",
        "    with torch.no_grad():\n",
        "        outputs = model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "        logits = outputs.logits\n",
        "        probabilities = torch.sigmoid(logits).squeeze()\n",
        "        return probabilities.cpu().numpy()\n",
        "\n",
        "# Example test data\n",
        "test_sentences = [\"Test sentence 1\", \"Test sentence 2\", ...]  # Replace with your sentences\n",
        "true_labels = [...]  # Replace with the true labels for the test data\n",
        "\n",
        "predicted_probs = [predict(sentence, model, tokenizer, device) for sentence in test_sentences]\n",
        "predicted_labels = [(prob > 0.5).astype(int) for prob in predicted_probs]\n"
      ],
      "metadata": {
        "id": "aBNBKFE7zWN6"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from torch.utils.data import DataLoader\n",
        "\n",
        "# Create a DataLoader for the test dataset\n",
        "test_loader = DataLoader(test_dataset_with_labels, batch_size=32, shuffle=False)\n",
        "\n",
        "# List to store predictions\n",
        "all_predictions = []\n",
        "\n",
        "# Predict\n",
        "model.eval()  # Put the model in evaluation mode\n",
        "with torch.no_grad():\n",
        "    for batch in test_loader:\n",
        "        input_ids = batch['input_ids'].to(device)\n",
        "        attention_mask = batch['attention_mask'].to(device)\n",
        "\n",
        "        outputs = model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "        logits = outputs.logits\n",
        "        predictions = torch.sigmoid(logits)\n",
        "        all_predictions.append(predictions)\n",
        "\n",
        "# Concatenate all predictions\n",
        "predictions = torch.cat(all_predictions, dim=0)\n"
      ],
      "metadata": {
        "id": "X0tJypWc0idw"
      },
      "execution_count": 65,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "threshold = 0.5\n",
        "predicted_labels = (predictions > threshold).cpu().numpy()\n",
        "\n"
      ],
      "metadata": {
        "id": "hythMNPv0sEh"
      },
      "execution_count": 76,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import classification_report\n",
        "\n",
        "\n",
        "true_labels = np.array([item['labels'].cpu().numpy() for item in test_dataset_with_labels])\n",
        "\n",
        "\n",
        "predicted_labels = (predictions > threshold).cpu().numpy()\n",
        "\n",
        "\n",
        "report = classification_report(true_labels, predicted_labels, target_names=['toxic', 'severe_toxic', 'obscene', 'threat', 'insult', 'identity_hate'])\n",
        "\n",
        "print(report)\n",
        "\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pDkpZU1w05C4",
        "outputId": "56878a1a-e012-479f-b582-43874389c649"
      },
      "execution_count": 81,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "               precision    recall  f1-score   support\n",
            "\n",
            "        toxic       0.50      0.92      0.65      6090\n",
            " severe_toxic       0.32      0.59      0.41       367\n",
            "      obscene       0.55      0.82      0.66      3691\n",
            "       threat       0.60      0.49      0.54       211\n",
            "       insult       0.64      0.74      0.69      3427\n",
            "identity_hate       0.79      0.41      0.54       712\n",
            "\n",
            "    micro avg       0.54      0.81      0.65     14498\n",
            "    macro avg       0.57      0.66      0.58     14498\n",
            " weighted avg       0.56      0.81      0.65     14498\n",
            "  samples avg       0.08      0.08      0.08     14498\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in samples with no predicted labels. Use `zero_division` parameter to control this behavior.\n",
            "  _warn_prf(average, modifier, msg_start, len(result))\n",
            "/usr/local/lib/python3.10/dist-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in samples with no true labels. Use `zero_division` parameter to control this behavior.\n",
            "  _warn_prf(average, modifier, msg_start, len(result))\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "2eJhCZIv1CdY"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}