TrainingAndEvaluationNotebook.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "zQz_5h_881S3"
   },
   "source": [
    "# Entrenamiento múltiple"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "gOFj999-b845"
   },
   "source": [
    "## Especificamos nuestra carpeta de trabajo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "!mkdir modelos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "867Kcr9-9XIL",
    "outputId": "9baa601f-d2d0-4bea-a636-b4ab31461ba4",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "total 8\r\n",
      "drwxr-xr-x  2 root root 4096 Mar 28 10:08 .\r\n",
      "drwxr-x--- 14 root root 4096 Mar 28 10:08 ..\r\n"
     ]
    }
   ],
   "source": [
    "workdir = '/tf/modelos/'\n",
    "workdir_datasets = '/tf/DATASETS/'\n",
    "!ls -la $workdir"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "oMyyJBHpLwgM"
   },
   "source": [
    "## Instalamos las librerías que necesitaremos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "gzjc_ohIL0ra",
    "outputId": "b7744cad-ade8-456e-ae9f-2dc7cdbc4fe8",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
      "\u001b[0m\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
      "\u001b[0m\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
      "\u001b[0m"
     ]
    }
   ],
   "source": [
    "!pip install -q tflite-model-maker\n",
    "!pip install -q tensorflow\n",
    "!pip install -q tensorflow-addons"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "_2gUf4AsNLrG"
   },
   "source": [
    "## Importamos las librerías que vamos a usar"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "OsLa8jY0NLTZ",
    "outputId": "222bda8b-7eb8-4de7-ea1a-eb0de614afac"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from numpy.random import RandomState\n",
    "import pandas as pd\n",
    "import os\n",
    "#from google.colab import files # Para descargar archivos desde Google Colab\n",
    "\n",
    "import tensorflow_addons as tfa\n",
    "\n",
    "from tflite_model_maker import model_spec\n",
    "from tflite_model_maker import text_classifier\n",
    "from tflite_model_maker.config import ExportFormat\n",
    "from tflite_model_maker.config import QuantizationConfig\n",
    "from tflite_model_maker.text_classifier import AverageWordVecSpec\n",
    "from tflite_model_maker.text_classifier import BertClassifierSpec\n",
    "from tflite_model_maker.text_classifier import MobileBertClassifierSpec\n",
    "from tflite_model_maker.text_classifier import DataLoader\n",
    "\n",
    "# Para ahorrarnos el tokenizer manual :)\n",
    "from tflite_support.task.text import NLClassifier\n",
    "from tflite_support.task.text import NLClassifierOptions\n",
    "import tflite_support\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import re\n",
    "import datetime\n",
    "\n",
    "import json\n",
    "import xlsxwriter\n",
    "from sklearn.metrics import accuracy_score\n",
    "from sklearn.metrics import log_loss\n",
    "from sklearn.metrics import recall_score\n",
    "from sklearn.metrics import precision_score\n",
    "from sklearn.metrics import classification_report\n",
    "\n",
    "import tensorflow as tf\n",
    "assert tf.__version__.startswith('2')\n",
    "tf.get_logger().setLevel('ERROR')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "MonfXku5953p"
   },
   "source": [
    "## Recogemos los datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "id": "Fp2q0hNX-I3M"
   },
   "outputs": [],
   "source": [
    "!curl -s https://raw.githubusercontent.com/Text-Analysis-Android-Apps/Custom-Text-Classification-on-Android-using-TF-Lite/master/train.csv > \"$workdir_datasets\"train_en.csv\n",
    "!curl -s https://raw.githubusercontent.com/Text-Analysis-Android-Apps/Custom-Text-Classification-on-Android-using-TF-Lite/master/test.csv > \"$workdir_datasets\"test_en.csv"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "90bJ-1P6Lit6"
   },
   "source": [
    "## Los importamos a Pandas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "tm9Zd_jHLkzi",
    "outputId": "62adc766-f58c-424d-d2cf-448b93bcb07d",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<ipython-input-13-f2749630dde0>:1: FutureWarning: The error_bad_lines argument has been deprecated and will be removed in a future version. Use on_bad_lines in the future.\n",
      "\n",
      "\n",
      "  df_train_en = pd.read_csv(workdir_datasets+'train_en.csv', error_bad_lines=False, engine=\"python\")\n",
      "<ipython-input-13-f2749630dde0>:2: FutureWarning: The error_bad_lines argument has been deprecated and will be removed in a future version. Use on_bad_lines in the future.\n",
      "\n",
      "\n",
      "  df_test_en = pd.read_csv(workdir_datasets+'test_en.csv', error_bad_lines=False, engine=\"python\")\n"
     ]
    }
   ],
   "source": [
    "df_train_en = pd.read_csv(workdir_datasets+'train_en.csv', error_bad_lines=False, engine=\"python\")\n",
    "df_test_en = pd.read_csv(workdir_datasets+'test_en.csv', error_bad_lines=False, engine=\"python\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "24GxsTooZVC4"
   },
   "source": [
    "## Procesamos el dataset inglés"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "id": "0tKXudKgYPz6"
   },
   "outputs": [],
   "source": [
    "df_test_en['label'] = df_test_en['label'].str.replace('Non-offensive', 'Non-Offensive')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "LfponKEBZcFH"
   },
   "source": [
    "### Vemos su tamaño"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "SwYNP76sY-Rn",
    "outputId": "46a3ef1f-2ed1-4927-b85c-6e99d89483ef"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "label\n",
       "Non-Offensive    350\n",
       "Offensive        350\n",
       "dtype: int64"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train_en.groupby('label').size()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "2T53buybZDtj",
    "outputId": "efe19dd7-e011-4aa8-8ea3-9f50f68498e0",
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "label\n",
       "Non-Offensive    150\n",
       "Offensive        150\n",
       "dtype: int64"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test_en.groupby('label').size()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "M5GRp8oVcEWs"
   },
   "source": [
    "## Exportamos los datasets procesados a csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "pC9cNC5GcIqU"
   },
   "outputs": [],
   "source": [
    "df_train_en.to_csv(workdir_datasets+'final_train_en.csv', index=True, sep='\\t')\n",
    "df_test_en.to_csv(workdir_datasets+'final_test_en.csv', index=True, sep='\\t')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "9MD_EqemVI5t"
   },
   "source": [
    "## Especificamos los idiomas y las opciones que se deben usar en un diccionario"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "id": "82jQdmhbVScd"
   },
   "outputs": [],
   "source": [
    "defaul_batch_size = 32\n",
    "opciones_modelos = [\n",
    "    {'lang': 'en', 'model': 'average_word_vec', 'epochs': 300, 'steps': None, 'batch_size': defaul_batch_size},\n",
    "    {'lang': 'en', 'model': 'mbert', 'epochs': 3, 'steps': None, 'batch_size': defaul_batch_size}\n",
    "    ]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "B2H8Rwy2WMCs"
   },
   "source": [
    "## Entrenamos los modelos especificados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "colab": {
     "background_save": true,
     "base_uri": "https://localhost:8080/"
    },
    "id": "b4K9A9BAazTr",
    "outputId": "3613b3ec-c647-4969-ef4b-b884e2a50328",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Entrenando modelo average_word_vec con dataset en:\n",
      "21/21 [==============================] - 1s 6ms/step - loss: 0.6932 - accuracy: 0.4955\n",
      "Epoch 2/2\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6922 - accuracy: 0.5670\n",
      "Epoch 3/3\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.6906 - accuracy: 0.6012\n",
      "Epoch 4/4\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6893 - accuracy: 0.6235\n",
      "Epoch 5/5\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6869 - accuracy: 0.6533\n",
      "Epoch 6/6\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6842 - accuracy: 0.6905\n",
      "Epoch 7/7\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6819 - accuracy: 0.7188\n",
      "Epoch 8/8\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6795 - accuracy: 0.7277\n",
      "Epoch 9/9\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6745 - accuracy: 0.7753\n",
      "Epoch 10/10\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6699 - accuracy: 0.8006\n",
      "Epoch 11/11\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6636 - accuracy: 0.7917\n",
      "Epoch 12/12\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6574 - accuracy: 0.7887\n",
      "Epoch 13/13\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6544 - accuracy: 0.8065\n",
      "Epoch 14/14\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6429 - accuracy: 0.8571\n",
      "Epoch 15/15\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6326 - accuracy: 0.8527\n",
      "Epoch 16/16\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6252 - accuracy: 0.8571\n",
      "Epoch 17/17\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6151 - accuracy: 0.8750\n",
      "Epoch 18/18\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.6029 - accuracy: 0.8839\n",
      "Epoch 19/19\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.5913 - accuracy: 0.8810\n",
      "Epoch 20/20\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.5827 - accuracy: 0.8914\n",
      "Epoch 21/21\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.5602 - accuracy: 0.9092\n",
      "Epoch 22/22\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.5492 - accuracy: 0.8943\n",
      "Epoch 23/23\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.5312 - accuracy: 0.9122\n",
      "Epoch 24/24\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.5193 - accuracy: 0.9167\n",
      "Epoch 25/25\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.5051 - accuracy: 0.9077\n",
      "Epoch 26/26\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.4902 - accuracy: 0.9137\n",
      "Epoch 27/27\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.4689 - accuracy: 0.9182\n",
      "Epoch 28/28\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.4535 - accuracy: 0.9196\n",
      "Epoch 29/29\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.4402 - accuracy: 0.9182\n",
      "Epoch 30/30\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.4189 - accuracy: 0.9241\n",
      "Epoch 31/31\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.4024 - accuracy: 0.9345\n",
      "Epoch 32/32\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.3884 - accuracy: 0.9271\n",
      "Epoch 33/33\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.3734 - accuracy: 0.9330\n",
      "Epoch 34/34\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.3596 - accuracy: 0.9330\n",
      "Epoch 35/35\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.3412 - accuracy: 0.9435\n",
      "Epoch 36/36\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.3228 - accuracy: 0.9330\n",
      "Epoch 37/37\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.3109 - accuracy: 0.9449\n",
      "Epoch 38/38\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.2943 - accuracy: 0.9539\n",
      "Epoch 39/39\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.2802 - accuracy: 0.9539\n",
      "Epoch 40/40\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.2688 - accuracy: 0.9494\n",
      "Epoch 41/41\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.2452 - accuracy: 0.9658\n",
      "Epoch 42/42\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.2424 - accuracy: 0.9479\n",
      "Epoch 43/43\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.2272 - accuracy: 0.9554\n",
      "Epoch 44/44\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.2107 - accuracy: 0.9673\n",
      "Epoch 45/45\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.2086 - accuracy: 0.9494\n",
      "Epoch 46/46\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.1978 - accuracy: 0.9628\n",
      "Epoch 47/47\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1876 - accuracy: 0.9658\n",
      "Epoch 48/48\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1827 - accuracy: 0.9554\n",
      "Epoch 49/49\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1654 - accuracy: 0.9673\n",
      "Epoch 50/50\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1636 - accuracy: 0.9628\n",
      "Epoch 51/51\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1465 - accuracy: 0.9673\n",
      "Epoch 52/52\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1327 - accuracy: 0.9792\n",
      "Epoch 53/53\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.1356 - accuracy: 0.9747\n",
      "Epoch 54/54\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.1280 - accuracy: 0.9807\n",
      "Epoch 55/55\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.1110 - accuracy: 0.9821\n",
      "Epoch 56/56\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1199 - accuracy: 0.9836\n",
      "Epoch 57/57\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.1160 - accuracy: 0.9717\n",
      "Epoch 58/58\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.1051 - accuracy: 0.9792\n",
      "Epoch 59/59\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.1000 - accuracy: 0.9777\n",
      "Epoch 60/60\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0984 - accuracy: 0.9836\n",
      "Epoch 61/61\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0944 - accuracy: 0.9851\n",
      "Epoch 62/62\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0881 - accuracy: 0.9866\n",
      "Epoch 63/63\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0807 - accuracy: 0.9881\n",
      "Epoch 64/64\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0720 - accuracy: 0.9881\n",
      "Epoch 65/65\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0773 - accuracy: 0.9821\n",
      "Epoch 66/66\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0752 - accuracy: 0.9881\n",
      "Epoch 67/67\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0627 - accuracy: 0.9940\n",
      "Epoch 68/68\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0663 - accuracy: 0.9851\n",
      "Epoch 69/69\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0579 - accuracy: 0.9911\n",
      "Epoch 70/70\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0582 - accuracy: 0.9881\n",
      "Epoch 71/71\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0525 - accuracy: 0.9896\n",
      "Epoch 72/72\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0504 - accuracy: 0.9881\n",
      "Epoch 73/73\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0470 - accuracy: 0.9911\n",
      "Epoch 74/74\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0503 - accuracy: 0.9896\n",
      "Epoch 75/75\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0478 - accuracy: 0.9911\n",
      "Epoch 76/76\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0366 - accuracy: 0.9940\n",
      "Epoch 77/77\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0343 - accuracy: 1.0000\n",
      "Epoch 78/78\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0340 - accuracy: 0.9955\n",
      "Epoch 79/79\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0394 - accuracy: 0.9940\n",
      "Epoch 80/80\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0378 - accuracy: 0.9896\n",
      "Epoch 81/81\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0327 - accuracy: 0.9955\n",
      "Epoch 82/82\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0313 - accuracy: 0.9940\n",
      "Epoch 83/83\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0284 - accuracy: 0.9955\n",
      "Epoch 84/84\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0271 - accuracy: 0.9970\n",
      "Epoch 85/85\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0307 - accuracy: 0.9970\n",
      "Epoch 86/86\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0264 - accuracy: 0.9955\n",
      "Epoch 87/87\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0303 - accuracy: 0.9940\n",
      "Epoch 88/88\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0251 - accuracy: 0.9970\n",
      "Epoch 89/89\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0199 - accuracy: 0.9970\n",
      "Epoch 90/90\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0255 - accuracy: 0.9955\n",
      "Epoch 91/91\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0228 - accuracy: 0.9955\n",
      "Epoch 92/92\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0212 - accuracy: 0.9985\n",
      "Epoch 93/93\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0170 - accuracy: 0.9970\n",
      "Epoch 94/94\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0163 - accuracy: 0.9985\n",
      "Epoch 95/95\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0166 - accuracy: 0.9970\n",
      "Epoch 96/96\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0152 - accuracy: 0.9970\n",
      "Epoch 97/97\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0176 - accuracy: 0.9985\n",
      "Epoch 98/98\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0128 - accuracy: 1.0000\n",
      "Epoch 99/99\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0138 - accuracy: 0.9970\n",
      "Epoch 100/100\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0136 - accuracy: 1.0000\n",
      "Epoch 101/101\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0109 - accuracy: 1.0000\n",
      "Epoch 102/102\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0131 - accuracy: 0.9970\n",
      "Epoch 103/103\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0107 - accuracy: 1.0000\n",
      "Epoch 104/104\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0085 - accuracy: 1.0000\n",
      "Epoch 105/105\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0102 - accuracy: 1.0000\n",
      "Epoch 106/106\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0106 - accuracy: 1.0000\n",
      "Epoch 107/107\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0070 - accuracy: 1.0000\n",
      "Epoch 108/108\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0092 - accuracy: 1.0000\n",
      "Epoch 109/109\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0103 - accuracy: 0.9985\n",
      "Epoch 110/110\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0064 - accuracy: 1.0000\n",
      "Epoch 111/111\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0104 - accuracy: 1.0000\n",
      "Epoch 112/112\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0075 - accuracy: 1.0000\n",
      "Epoch 113/113\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0076 - accuracy: 0.9985\n",
      "Epoch 114/114\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0078 - accuracy: 1.0000\n",
      "Epoch 115/115\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0105 - accuracy: 0.9955\n",
      "Epoch 116/116\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0053 - accuracy: 1.0000\n",
      "Epoch 117/117\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0070 - accuracy: 1.0000\n",
      "Epoch 118/118\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0047 - accuracy: 1.0000\n",
      "Epoch 119/119\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0060 - accuracy: 0.9985\n",
      "Epoch 120/120\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0056 - accuracy: 0.9985\n",
      "Epoch 121/121\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0066 - accuracy: 1.0000\n",
      "Epoch 122/122\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0026 - accuracy: 1.0000\n",
      "Epoch 123/123\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0053 - accuracy: 1.0000\n",
      "Epoch 124/124\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0056 - accuracy: 1.0000\n",
      "Epoch 125/125\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0063 - accuracy: 0.9985\n",
      "Epoch 126/126\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0048 - accuracy: 1.0000\n",
      "Epoch 127/127\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0039 - accuracy: 1.0000\n",
      "Epoch 128/128\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0038 - accuracy: 1.0000\n",
      "Epoch 129/129\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0058 - accuracy: 0.9985\n",
      "Epoch 130/130\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0054 - accuracy: 0.9970\n",
      "Epoch 131/131\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0038 - accuracy: 1.0000\n",
      "Epoch 132/132\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0027 - accuracy: 1.0000\n",
      "Epoch 133/133\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0026 - accuracy: 1.0000\n",
      "Epoch 134/134\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0027 - accuracy: 1.0000\n",
      "Epoch 135/135\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0039 - accuracy: 1.0000\n",
      "Epoch 136/136\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0028 - accuracy: 1.0000\n",
      "Epoch 137/137\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0059 - accuracy: 0.9985\n",
      "Epoch 138/138\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0044 - accuracy: 0.9985\n",
      "Epoch 139/139\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0031 - accuracy: 1.0000\n",
      "Epoch 140/140\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0038 - accuracy: 1.0000\n",
      "Epoch 141/141\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0042 - accuracy: 0.9985\n",
      "Epoch 142/142\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0033 - accuracy: 1.0000\n",
      "Epoch 143/143\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0025 - accuracy: 1.0000\n",
      "Epoch 144/144\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0026 - accuracy: 1.0000\n",
      "Epoch 145/145\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0037 - accuracy: 0.9985\n",
      "Epoch 146/146\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0011 - accuracy: 1.0000\n",
      "Epoch 147/147\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0024 - accuracy: 1.0000\n",
      "Epoch 148/148\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0021 - accuracy: 1.0000\n",
      "Epoch 149/149\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0024 - accuracy: 1.0000\n",
      "Epoch 150/150\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0025 - accuracy: 1.0000\n",
      "Epoch 151/151\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0014 - accuracy: 1.0000\n",
      "Epoch 152/152\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.0024 - accuracy: 1.0000\n",
      "Epoch 153/153\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0029 - accuracy: 0.9985\n",
      "Epoch 154/154\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.0026 - accuracy: 1.0000\n",
      "Epoch 155/155\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.0022 - accuracy: 1.0000\n",
      "Epoch 156/156\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.8449e-04 - accuracy: 1.0000\n",
      "Epoch 157/157\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 1.0000\n",
      "Epoch 158/158\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0017 - accuracy: 1.0000\n",
      "Epoch 159/159\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0012 - accuracy: 1.0000\n",
      "Epoch 160/160\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0012 - accuracy: 1.0000\n",
      "Epoch 161/161\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0026 - accuracy: 1.0000\n",
      "Epoch 162/162\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0021 - accuracy: 1.0000\n",
      "Epoch 163/163\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.0019 - accuracy: 1.0000\n",
      "Epoch 164/164\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 1.0000\n",
      "Epoch 165/165\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0016 - accuracy: 1.0000\n",
      "Epoch 166/166\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.7171e-04 - accuracy: 1.0000\n",
      "Epoch 167/167\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0011 - accuracy: 1.0000\n",
      "Epoch 168/168\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0012 - accuracy: 1.0000\n",
      "Epoch 169/169\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0016 - accuracy: 1.0000\n",
      "Epoch 170/170\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 6.2380e-04 - accuracy: 1.0000\n",
      "Epoch 171/171\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 5.3922e-04 - accuracy: 1.0000\n",
      "Epoch 172/172\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0014 - accuracy: 1.0000\n",
      "Epoch 173/173\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 9.4344e-04 - accuracy: 1.0000\n",
      "Epoch 174/174\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0019 - accuracy: 1.0000\n",
      "Epoch 175/175\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.7529e-04 - accuracy: 1.0000\n",
      "Epoch 176/176\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 8.3292e-04 - accuracy: 1.0000\n",
      "Epoch 177/177\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0026 - accuracy: 1.0000\n",
      "Epoch 178/178\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0017 - accuracy: 1.0000\n",
      "Epoch 179/179\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0019 - accuracy: 1.0000\n",
      "Epoch 180/180\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0010 - accuracy: 1.0000\n",
      "Epoch 181/181\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.4881e-04 - accuracy: 1.0000\n",
      "Epoch 182/182\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 1.0000\n",
      "Epoch 183/183\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0014 - accuracy: 1.0000\n",
      "Epoch 184/184\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.1012e-04 - accuracy: 1.0000\n",
      "Epoch 185/185\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.8455e-04 - accuracy: 1.0000\n",
      "Epoch 186/186\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.9330e-04 - accuracy: 1.0000\n",
      "Epoch 187/187\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0011 - accuracy: 1.0000\n",
      "Epoch 188/188\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0011 - accuracy: 1.0000\n",
      "Epoch 189/189\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.8837e-04 - accuracy: 1.0000\n",
      "Epoch 190/190\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0013 - accuracy: 1.0000\n",
      "Epoch 191/191\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 1.0000\n",
      "Epoch 192/192\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0019 - accuracy: 1.0000\n",
      "Epoch 193/193\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0012 - accuracy: 1.0000\n",
      "Epoch 194/194\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.0702e-04 - accuracy: 1.0000\n",
      "Epoch 195/195\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0013 - accuracy: 1.0000\n",
      "Epoch 196/196\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 3.8033e-04 - accuracy: 1.0000\n",
      "Epoch 197/197\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 6.1590e-04 - accuracy: 1.0000\n",
      "Epoch 198/198\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 8.2647e-04 - accuracy: 1.0000\n",
      "Epoch 199/199\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 4.2215e-04 - accuracy: 1.0000\n",
      "Epoch 200/200\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0024 - accuracy: 0.9985\n",
      "Epoch 201/201\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 6.3931e-04 - accuracy: 1.0000\n",
      "Epoch 202/202\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.5687e-04 - accuracy: 1.0000\n",
      "Epoch 203/203\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.6821e-04 - accuracy: 1.0000\n",
      "Epoch 204/204\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.6737e-04 - accuracy: 1.0000\n",
      "Epoch 205/205\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.8828e-04 - accuracy: 1.0000\n",
      "Epoch 206/206\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0023 - accuracy: 0.9985\n",
      "Epoch 207/207\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 2.6204e-04 - accuracy: 1.0000\n",
      "Epoch 208/208\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.6244e-04 - accuracy: 1.0000\n",
      "Epoch 209/209\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.5107e-04 - accuracy: 1.0000\n",
      "Epoch 210/210\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 9.3042e-04 - accuracy: 1.0000\n",
      "Epoch 211/211\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 9.0069e-04 - accuracy: 1.0000\n",
      "Epoch 212/212\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.0344e-04 - accuracy: 1.0000\n",
      "Epoch 213/213\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 7.1527e-04 - accuracy: 1.0000\n",
      "Epoch 214/214\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 4.7694e-04 - accuracy: 1.0000\n",
      "Epoch 215/215\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.2860e-04 - accuracy: 1.0000\n",
      "Epoch 216/216\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.3977e-04 - accuracy: 1.0000\n",
      "Epoch 217/217\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.4533e-04 - accuracy: 1.0000\n",
      "Epoch 218/218\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.2150e-04 - accuracy: 1.0000\n",
      "Epoch 219/219\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 3.9569e-04 - accuracy: 1.0000\n",
      "Epoch 220/220\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 0.9985\n",
      "Epoch 221/221\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 8.0992e-04 - accuracy: 1.0000\n",
      "Epoch 222/222\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.5306e-04 - accuracy: 1.0000\n",
      "Epoch 223/223\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 7.6684e-04 - accuracy: 1.0000\n",
      "Epoch 224/224\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0015 - accuracy: 0.9985\n",
      "Epoch 225/225\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 3.2466e-04 - accuracy: 1.0000\n",
      "Epoch 226/226\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.1183e-04 - accuracy: 1.0000\n",
      "Epoch 227/227\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 3.3468e-04 - accuracy: 1.0000\n",
      "Epoch 228/228\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.4397e-04 - accuracy: 1.0000\n",
      "Epoch 229/229\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.7127e-04 - accuracy: 1.0000\n",
      "Epoch 230/230\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0043 - accuracy: 0.9985\n",
      "Epoch 231/231\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.3901e-04 - accuracy: 1.0000\n",
      "Epoch 232/232\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.7526e-04 - accuracy: 1.0000\n",
      "Epoch 233/233\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.2735e-04 - accuracy: 1.0000\n",
      "Epoch 234/234\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0010 - accuracy: 1.0000\n",
      "Epoch 235/235\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.6294e-04 - accuracy: 1.0000\n",
      "Epoch 236/236\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.3835e-04 - accuracy: 1.0000\n",
      "Epoch 237/237\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 1.0000\n",
      "Epoch 238/238\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.1050e-04 - accuracy: 1.0000\n",
      "Epoch 239/239\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 8.7335e-05 - accuracy: 1.0000\n",
      "Epoch 240/240\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 9.3622e-05 - accuracy: 1.0000\n",
      "Epoch 241/241\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.7042e-04 - accuracy: 1.0000\n",
      "Epoch 242/242\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.0543e-04 - accuracy: 1.0000\n",
      "Epoch 243/243\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.9886e-04 - accuracy: 1.0000\n",
      "Epoch 244/244\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.0255e-04 - accuracy: 1.0000\n",
      "Epoch 245/245\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.7686e-04 - accuracy: 1.0000\n",
      "Epoch 246/246\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.2066e-04 - accuracy: 1.0000\n",
      "Epoch 247/247\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.8501e-04 - accuracy: 1.0000\n",
      "Epoch 248/248\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.4620e-04 - accuracy: 1.0000\n",
      "Epoch 249/249\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.5121e-04 - accuracy: 1.0000\n",
      "Epoch 250/250\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.0640e-05 - accuracy: 1.0000\n",
      "Epoch 251/251\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.2554e-04 - accuracy: 1.0000\n",
      "Epoch 252/252\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.1833e-05 - accuracy: 1.0000\n",
      "Epoch 253/253\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0015 - accuracy: 1.0000\n",
      "Epoch 254/254\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 1.7319e-04 - accuracy: 1.0000\n",
      "Epoch 255/255\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 5.1789e-04 - accuracy: 1.0000\n",
      "Epoch 256/256\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.0018 - accuracy: 0.9985\n",
      "Epoch 257/257\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.9330e-04 - accuracy: 1.0000\n",
      "Epoch 258/258\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.3293e-04 - accuracy: 1.0000\n",
      "Epoch 259/259\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.5399e-04 - accuracy: 1.0000\n",
      "Epoch 260/260\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.6352e-05 - accuracy: 1.0000\n",
      "Epoch 261/261\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.8770e-05 - accuracy: 1.0000\n",
      "Epoch 262/262\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 8.0985e-04 - accuracy: 1.0000\n",
      "Epoch 263/263\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.9403e-04 - accuracy: 1.0000\n",
      "Epoch 264/264\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 5.2386e-04 - accuracy: 1.0000\n",
      "Epoch 265/265\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.8223e-04 - accuracy: 1.0000\n",
      "Epoch 266/266\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.5593e-04 - accuracy: 1.0000\n",
      "Epoch 267/267\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.8700e-05 - accuracy: 1.0000\n",
      "Epoch 268/268\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.6166e-05 - accuracy: 1.0000\n",
      "Epoch 269/269\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.0816e-04 - accuracy: 1.0000\n",
      "Epoch 270/270\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.8258e-04 - accuracy: 1.0000\n",
      "Epoch 271/271\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 9.9918e-05 - accuracy: 1.0000\n",
      "Epoch 272/272\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.1322e-05 - accuracy: 1.0000\n",
      "Epoch 273/273\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.4368e-05 - accuracy: 1.0000\n",
      "Epoch 274/274\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.0142e-04 - accuracy: 1.0000\n",
      "Epoch 275/275\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 0.0011 - accuracy: 1.0000\n",
      "Epoch 276/276\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.0014 - accuracy: 1.0000\n",
      "Epoch 277/277\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 4.1496e-04 - accuracy: 1.0000\n",
      "Epoch 278/278\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.8437e-04 - accuracy: 1.0000\n",
      "Epoch 279/279\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.3748e-04 - accuracy: 1.0000\n",
      "Epoch 280/280\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 7.1278e-05 - accuracy: 1.0000\n",
      "Epoch 281/281\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 5.5471e-05 - accuracy: 1.0000\n",
      "Epoch 282/282\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.1824e-04 - accuracy: 1.0000\n",
      "Epoch 283/283\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.5574e-04 - accuracy: 1.0000\n",
      "Epoch 284/284\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.9899e-04 - accuracy: 1.0000\n",
      "Epoch 285/285\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.7663e-05 - accuracy: 1.0000\n",
      "Epoch 286/286\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.4625e-04 - accuracy: 1.0000\n",
      "Epoch 287/287\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 1.3161e-05 - accuracy: 1.0000\n",
      "Epoch 288/288\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.4204e-05 - accuracy: 1.0000\n",
      "Epoch 289/289\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 8.3613e-05 - accuracy: 1.0000\n",
      "Epoch 290/290\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.5377e-05 - accuracy: 1.0000\n",
      "Epoch 291/291\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.4995e-05 - accuracy: 1.0000\n",
      "Epoch 292/292\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.6224e-04 - accuracy: 1.0000\n",
      "Epoch 293/293\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.8864e-04 - accuracy: 1.0000\n",
      "Epoch 294/294\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.1062e-04 - accuracy: 1.0000\n",
      "Epoch 295/295\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 6.7111e-04 - accuracy: 1.0000\n",
      "Epoch 296/296\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.0021e-04 - accuracy: 1.0000\n",
      "Epoch 297/297\n",
      "21/21 [==============================] - 0s 7ms/step - loss: 3.4161e-05 - accuracy: 1.0000\n",
      "Epoch 298/298\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 4.6227e-04 - accuracy: 1.0000\n",
      "Epoch 299/299\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.8682e-05 - accuracy: 1.0000\n",
      "Epoch 300/300\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 5.6392e-04 - accuracy: 1.0000\n",
      "Mostrando un gráfico del modelo:\n",
      "Model: \"sequential_5\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " embedding_5 (Embedding)     (None, 256, 16)           38064     \n",
      "                                                                 \n",
      " global_average_pooling1d_5   (None, 16)               0         \n",
      " (GlobalAveragePooling1D)                                        \n",
      "                                                                 \n",
      " dense_10 (Dense)            (None, 16)                272       \n",
      "                                                                 \n",
      " dropout_7 (Dropout)         (None, 16)                0         \n",
      "                                                                 \n",
      " dense_11 (Dense)            (None, 2)                 34        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 38,370\n",
      "Trainable params: 38,370\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Calculando accuracy y loss final:\n",
      "10/10 [==============================] - 0s 9ms/step - loss: 0.1138 - accuracy: 0.9600\n",
      "Exportando modelo...\n",
      "Realizando predicciones y calculando el tiempo medio de inferencia...\n",
      "Las predicciones se han realizado en 0.147519 segundos.\n",
      "Se ha generado un modelo de tipo average_word_vec, idioma en, con 300 epochs, None steps y batch_size 32 con una accuracy de 0.9599999785423279 y un loss de 0.11383674293756485\n",
      "Guardando las predicciones en model_average_word_vec_en-y_pred-train.json...\n",
      "¡Modelo listo!\n",
      "\n",
      "\n",
      "Entrenando modelo mbert con dataset en:\n",
      "21/21 [==============================] - 65s 297ms/step - loss: 1.7974 - test_accuracy: 0.7068\n",
      "Epoch 2/2\n",
      "21/21 [==============================] - 11s 372ms/step - loss: 0.1989 - test_accuracy: 0.9420\n",
      "Epoch 3/3\n",
      "21/21 [==============================] - 10s 337ms/step - loss: 0.0874 - test_accuracy: 0.9702\n",
      "Mostrando un gráfico del modelo:\n",
      "Model: \"model_2\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_word_ids (InputLayer)    [(None, 128)]        0           []                               \n",
      "                                                                                                  \n",
      " input_mask (InputLayer)        [(None, 128)]        0           []                               \n",
      "                                                                                                  \n",
      " input_type_ids (InputLayer)    [(None, 128)]        0           []                               \n",
      "                                                                                                  \n",
      " hub_keras_layer_v1v2_2 (HubKer  (None, 512)         24581888    ['input_word_ids[0][0]',         \n",
      " asLayerV1V2)                                                     'input_mask[0][0]',             \n",
      "                                                                  'input_type_ids[0][0]']         \n",
      "                                                                                                  \n",
      " dropout_8 (Dropout)            (None, 512)          0           ['hub_keras_layer_v1v2_2[0][0]'] \n",
      "                                                                                                  \n",
      " output (Dense)                 (None, 2)            1026        ['dropout_8[0][0]']              \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 24,582,914\n",
      "Trainable params: 24,582,914\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "Calculando accuracy y loss final:\n",
      "10/10 [==============================] - 7s 75ms/step - loss: 0.1282 - test_accuracy: 0.9633\n",
      "Exportando modelo...\n",
      "Realizando predicciones y calculando el tiempo medio de inferencia...\n",
      "Las predicciones se han realizado en 6.891475 segundos.\n",
      "Se ha generado un modelo de tipo mbert, idioma en, con 3 epochs, None steps y batch_size 32 con una accuracy de 0.9633333086967468 y un loss de 0.12815383076667786\n",
      "Guardando las predicciones en model_mbert_en-y_pred-train.json...\n",
      "¡Modelo listo!\n",
      "\n",
      "¡Se ha finalizado el entrenamiento de todos los modelos con éxito!\n",
      "Tiempo total de entrenamiento: 242.219835 segundos.\n"
     ]
    }
   ],
   "source": [
    "hora_comienzo = datetime.datetime.now()\n",
    "datos_modelos_generados = []\n",
    "y_preds = []\n",
    "for modelo in opciones_modelos:\n",
    "  model_name = \"model_\" + modelo['model'] + \"_\" + modelo['lang']\n",
    "  print(\"\\n\\nEntrenando modelo \"+modelo['model']+\" con dataset \"+modelo['lang']+\":\")\n",
    "  if (modelo['model'] == 'average_word_vec'):\n",
    "    spec = model_spec.get('average_word_vec')\n",
    "  elif (modelo['model'] == 'bert'):\n",
    "    spec = model_spec.get('bert_classifier')\n",
    "  elif (modelo['model'] == 'mbert'):\n",
    "    spec = model_spec.get('mobilebert_classifier')\n",
    "    #spec.seq_len = 256\n",
    "\n",
    "  train_data = DataLoader.from_csv(\n",
    "        filename=workdir_datasets+'final_train_'+modelo['lang']+'.csv',\n",
    "        delimiter='\\t',\n",
    "        text_column='sentence',\n",
    "        label_column='label',\n",
    "        model_spec=spec,\n",
    "        is_training=True)\n",
    "\n",
    "  test_data = DataLoader.from_csv(\n",
    "        filename=workdir_datasets+'final_test_'+modelo['lang']+'.csv',\n",
    "        delimiter='\\t',\n",
    "        text_column='sentence',\n",
    "        label_column='label',\n",
    "        model_spec=spec,\n",
    "        is_training=False)\n",
    "  modelo_actual = text_classifier.create(train_data, model_spec=spec, epochs=modelo['epochs'], batch_size=modelo['batch_size'], steps_per_epoch=modelo['steps'])\n",
    "\n",
    "  # Gráfico del modelo\n",
    "  print(\"Mostrando un gráfico del modelo:\")\n",
    "  modelo_actual.summary()\n",
    "  # Pequeña evaluación\n",
    "  print(\"Calculando accuracy y loss final:\")\n",
    "  loss, acc = modelo_actual.evaluate(test_data)\n",
    "\n",
    "  #modelos_generados.append(modelo_actual)\n",
    "  print(\"Exportando modelo...\")\n",
    "  # Se exporta el modelo\n",
    "  modelo_actual.export(export_dir=workdir, tflite_filename=model_name+'.tflite')\n",
    "  print(\"Realizando predicciones y calculando el tiempo medio de inferencia...\")\n",
    "\n",
    "  # Calculamos el tiempo de inferencia de todas las predicciones y lo almacenamos\n",
    "  tiempo_inicio = datetime.datetime.now()\n",
    "  y_pred = modelo_actual.predict_top_k(test_data, k=1)\n",
    "  tiempo_fin = datetime.datetime.now()\n",
    "  tiempo_inferencia_total = (tiempo_fin - tiempo_inicio).total_seconds()\n",
    "  print(\"Las predicciones se han realizado en \"+str(tiempo_inferencia_total)+\" segundos.\")\n",
    "  y_preds.append(y_pred)\n",
    "\n",
    "  # Backup de los datos básicos del modelo:\n",
    "  datos_modelo = {\"nombre\": model_name, 'tipo': modelo['model'], 'lang': modelo['lang'], 'epochs': modelo['epochs'], 'batch_size': modelo['batch_size'], 'steps': modelo['steps'], 'acc': acc, 'loss': loss, 'tiempo de inferencia total': tiempo_inferencia_total}\n",
    "  # Backup de los datos del modelo\n",
    "  with open(workdir+model_name+\"-datos.json\", \"w\") as f:\n",
    "      json.dump(datos_modelo, f)\n",
    "\n",
    "\n",
    "  datos_modelos_generados.append(datos_modelo)\n",
    "  print(\"Se ha generado un modelo de tipo \"+datos_modelo['tipo']+\", idioma \"+datos_modelo['lang']+\", con \"+str(datos_modelo['epochs'])+\" epochs, \"+str(datos_modelo['steps'])+\" steps y batch_size \"+str(datos_modelo['batch_size'])+\" con una accuracy de \"+str(datos_modelo['acc'])+\" y un loss de \"+str(datos_modelo['loss']))\n",
    "\n",
    "  # Backup de los datos predichos\n",
    "  print(\"Guardando las predicciones en \"+model_name+\"-y_pred-train.json...\")\n",
    "  y_pred_aux = []\n",
    "  for i in range(len(y_pred)):\n",
    "    y_pred_aux.append((y_pred[i][0][0], float(y_pred[i][0][1])))\n",
    "\n",
    "  with open(workdir+model_name+\"-y_pred-train.json\", \"w\") as f:\n",
    "      json.dump(y_pred_aux, f)\n",
    "  print(\"¡Modelo listo!\")\n",
    "\n",
    "print(\"\\n¡Se ha finalizado el entrenamiento de todos los modelos con éxito!\")\n",
    "hora_final = datetime.datetime.now()\n",
    "tiempo_inferencia_total = (hora_final - hora_comienzo).total_seconds()\n",
    "print(\"Tiempo total de entrenamiento: \"+str(tiempo_inferencia_total)+\" segundos.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "6By99zIWs8uV"
   },
   "source": [
    "# Evaluación de los modelos"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Las dos siguientes celdas generan distintas métricas de cada uno de los modelos y se almacenan en un archivo .json específico de cada modelo. Posteriormente, se unifican todos los datos obtenidos en una tabla en Markdown almacenada en el fichero README.md (para su reresentación en GitHub/GitLab), y en el fichero EvaluacionModelos.xlsx para su representación en Excel."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "def etiquetaMarkdown(texto, separador=False):\n",
    "  if separador:\n",
    "    return '-' * (len(texto) + 2) + '|'\n",
    "  else:\n",
    "    return ' ' + str(texto).replace('\\n', '<br>') + ' |'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "id": "dcE9oEgQs-pG"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluación del modelo model_average_word_vec_en: \n",
      "SKLearn accuracy score\n",
      "0.96\n",
      "SKLearn log loss\n",
      "1.3815617170954972\n",
      "SKLearn precision score (micro)\n",
      "0.96\n",
      "SKLearn precision score (macro)\n",
      "0.9603273438889877\n",
      "SKLearn recall score (micro)\n",
      "0.96\n",
      "SKLearn recall score (macro)\n",
      "0.96\n",
      "TFAddons recall (micro)\n",
      "0.96\n",
      "TFAddons F1 Score (categorias)\n",
      "[0.9605263 0.9594595]\n",
      "TFAddons F1 Score (micro)\n",
      "0.96\n",
      "TFAddons F1 Score (macro)\n",
      "0.9599929\n",
      "SKLearn classification report\n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "Non-Offensive       0.95      0.97      0.96       150\n",
      "    Offensive       0.97      0.95      0.96       150\n",
      "\n",
      "     accuracy                           0.96       300\n",
      "    macro avg       0.96      0.96      0.96       300\n",
      " weighted avg       0.96      0.96      0.96       300\n",
      "\n",
      "{'SKLearn accuracy score': '0.96', 'SKLearn log loss': '1.3815617170954972', 'SKLearn precision score (micro)': '0.96', 'SKLearn precision score (macro)': '0.9603273438889877', 'SKLearn recall score (micro)': '0.96', 'SKLearn recall score (macro)': '0.96', 'TFAddons recall (micro)': '0.96', 'TFAddons F1 Score (categorias)': '[0.9605263 0.9594595]', 'TFAddons F1 Score (micro)': '0.96', 'TFAddons F1 Score (macro)': '0.9599929', 'SKLearn classification report': '               precision    recall  f1-score   support\\n\\nNon-Offensive       0.95      0.97      0.96       150\\n    Offensive       0.97      0.95      0.96       150\\n\\n     accuracy                           0.96       300\\n    macro avg       0.96      0.96      0.96       300\\n weighted avg       0.96      0.96      0.96       300\\n'}\n",
      "Evaluación del modelo model_mbert_en: \n",
      "SKLearn accuracy score\n",
      "0.9633333333333334\n",
      "SKLearn log loss\n",
      "1.2664431237448643\n",
      "SKLearn precision score (micro)\n",
      "0.9633333333333334\n",
      "SKLearn precision score (macro)\n",
      "0.9638487208008899\n",
      "SKLearn recall score (micro)\n",
      "0.9633333333333334\n",
      "SKLearn recall score (macro)\n",
      "0.9633333333333334\n",
      "TFAddons recall (micro)\n",
      "0.9633333\n",
      "TFAddons F1 Score (categorias)\n",
      "[0.9627119  0.96393436]\n",
      "TFAddons F1 Score (micro)\n",
      "0.9633333\n",
      "TFAddons F1 Score (macro)\n",
      "0.9633231\n",
      "SKLearn classification report\n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "Non-Offensive       0.98      0.95      0.96       150\n",
      "    Offensive       0.95      0.98      0.96       150\n",
      "\n",
      "     accuracy                           0.96       300\n",
      "    macro avg       0.96      0.96      0.96       300\n",
      " weighted avg       0.96      0.96      0.96       300\n",
      "\n",
      "{'SKLearn accuracy score': '0.9633333333333334', 'SKLearn log loss': '1.2664431237448643', 'SKLearn precision score (micro)': '0.9633333333333334', 'SKLearn precision score (macro)': '0.9638487208008899', 'SKLearn recall score (micro)': '0.9633333333333334', 'SKLearn recall score (macro)': '0.9633333333333334', 'TFAddons recall (micro)': '0.9633333', 'TFAddons F1 Score (categorias)': '[0.9627119  0.96393436]', 'TFAddons F1 Score (micro)': '0.9633333', 'TFAddons F1 Score (macro)': '0.9633231', 'SKLearn classification report': '               precision    recall  f1-score   support\\n\\nNon-Offensive       0.98      0.95      0.96       150\\n    Offensive       0.95      0.98      0.96       150\\n\\n     accuracy                           0.96       300\\n    macro avg       0.96      0.96      0.96       300\\n weighted avg       0.96      0.96      0.96       300\\n'}\n"
     ]
    }
   ],
   "source": [
    "workbook = xlsxwriter.Workbook(workdir+'EvaluacionModelos.xlsx')\n",
    "worksheet = workbook.add_worksheet()\n",
    "bold = workbook.add_format({'bold': True})\n",
    "markdown_headers = '|'\n",
    "markdown_separator = '|'\n",
    "markdown_final = ''\n",
    "cabeceras = ['Modelo', 'Idioma', 'Tamaño Dataset entrenamiento', 'Tamaño dataset prueba',\n",
    "             'EPOCHs', 'Steps/EPOCH', 'batch_size', 'Loss', 'Accuracy', 'SKLearn accuracy score', 'SKLearn log loss'\n",
    "             'SKLearn precision score (micro)', 'SKLearn precision score (macro)', 'SKLearn recall score (micro)',\n",
    "             'SKLearn recall score (macro)', 'TensorFlowAddons recall (micro)', 'TensorFlowAddons F1 Score (categorias)',\n",
    "             'TensorFlowAddons F1 Score (micro)', 'TensorFlowAddons F1 Score (macro)', 'SKLearn classification report',\n",
    "             'Tiempo de medio de inferencia (s)']\n",
    "for columna in range(len(cabeceras)):\n",
    "  worksheet.write(0, columna, cabeceras[columna], bold)\n",
    "  markdown_headers += etiquetaMarkdown(cabeceras[columna])\n",
    "  markdown_separator += etiquetaMarkdown(cabeceras[columna], True)\n",
    "\n",
    "markdown_final = markdown_headers + '\\n' + markdown_separator + '\\n'\n",
    "\n",
    "i = 1\n",
    "for modelo in opciones_modelos:\n",
    "  markdown_modelo = '|'\n",
    "  model_name = \"model_\" + modelo['model'] + \"_\" + modelo['lang']\n",
    "  backup = {}\n",
    "  print(\"Evaluación del modelo \"+model_name+\": \")\n",
    "\n",
    "  # Introducimos los valores básicos del modelo\n",
    "  # Modelo\n",
    "  if (modelo['model'] == 'average_word_vec'):\n",
    "    worksheet.write(i, 0, 'Average_word_vector')\n",
    "    markdown_modelo += etiquetaMarkdown('Average_word_vector')\n",
    "  elif (modelo['model'] == 'bert'):\n",
    "    worksheet.write(i, 0, 'BERT')\n",
    "    markdown_modelo += etiquetaMarkdown('BERT')\n",
    "  elif (modelo['model'] == 'mbert'):\n",
    "    worksheet.write(i, 0, 'MobileBERT')\n",
    "    markdown_modelo += etiquetaMarkdown('MobileBERT')\n",
    "\n",
    "  # Idioma\n",
    "  if (modelo['lang'] == 'en'):\n",
    "    worksheet.write(i, 1, 'Inglés')\n",
    "    markdown_modelo += etiquetaMarkdown('Inglés')\n",
    "    worksheet.write(i, 2, '700 (balanceado)')\n",
    "    markdown_modelo += etiquetaMarkdown('700 (balanceado)')\n",
    "    worksheet.write(i, 3, '300 (balanceado)')\n",
    "    markdown_modelo += etiquetaMarkdown('300 (balanceado)')\n",
    "    df_test = df_test_en\n",
    "\n",
    "  # EPOCHS\n",
    "  worksheet.write(i, 4, modelo['epochs'])\n",
    "  markdown_modelo += etiquetaMarkdown(modelo['epochs'])\n",
    "  # Steps\n",
    "  worksheet.write(i, 5, modelo['steps'])\n",
    "  markdown_modelo += etiquetaMarkdown(modelo['steps'])\n",
    "  # batch_size\n",
    "  worksheet.write(i, 6, modelo['batch_size'])\n",
    "  markdown_modelo += etiquetaMarkdown(modelo['batch_size'])\n",
    "\n",
    "  # Accuracy y loss\n",
    "  with open(workdir+model_name+\"-datos.json\", \"r\") as f:\n",
    "    datos = json.load(f)\n",
    "    worksheet.write(i, 7, datos['loss'])\n",
    "    markdown_modelo += etiquetaMarkdown(datos['loss'])\n",
    "    worksheet.write(i, 8, datos['acc'])\n",
    "    markdown_modelo += etiquetaMarkdown(datos['acc'])\n",
    "\n",
    "\n",
    "  # Preparamos los datos para la evaluación\n",
    "  # Restauramos el y_pred guardado anteriormente\n",
    "  with open(workdir+model_name+\"-y_pred-train.json\", \"r\") as f:\n",
    "      y_pred = json.load(f)\n",
    "\n",
    "  #Conversión de array de arrays a array de paréntesis\n",
    "  y_pred_aux = []\n",
    "  for pred in y_pred:\n",
    "    y_pred_aux.append([(pred[0], float(pred[1]))])\n",
    "  y_pred = y_pred_aux\n",
    "\n",
    "  backup = {}\n",
    "\n",
    "  # SKLearn\n",
    "\n",
    "  # Creamos una matriz que represente en una dupla la etiqueta asignada a cada dato (con un 1) de los datos para test\n",
    "  y_true_sk = []\n",
    "  for label in df_test['label']:\n",
    "    if label == 'Non-Offensive':\n",
    "      y_true_sk.append(0)\n",
    "    else:\n",
    "      y_true_sk.append(1)\n",
    "\n",
    "  y_true_sk = np.array(y_true_sk, np.int32)\n",
    "\n",
    "  y_pred_sk = y_pred\n",
    "  # Creamos otra matriz que represente en una dupla la confianza de que un dato tiene cada etiqueta a partir de las predicciones hechas con los datos para test\n",
    "  y_pred_aux = []\n",
    "  for obj in y_pred_sk:\n",
    "    obj = obj[0]\n",
    "    if obj[0] == 'Non-Offensive':\n",
    "      y_pred_aux.append(0)\n",
    "    else:\n",
    "      y_pred_aux.append(1)\n",
    "\n",
    "  y_pred_sk = np.array(y_pred_aux, np.int32)\n",
    "\n",
    "\n",
    "\n",
    "  # TFAdditions\n",
    "\n",
    "  # Creamos una matriz que represente en una dupla la etiqueta asignada a cada dato (con un 1) de los datos para test\n",
    "  y_true = []\n",
    "  for label in df_test['label']:\n",
    "    if label == 'Non-Offensive':\n",
    "      y_true.append([1, 0])\n",
    "    else:\n",
    "      y_true.append([0, 1])\n",
    "\n",
    "  y_true = np.array(y_true, np.int32)\n",
    "\n",
    "  # Creamos otra matriz que represente en una dupla la confianza de que un dato tiene cada etiqueta a partir de las predicciones hechas con los datos para test\n",
    "  y_pred_aux = []\n",
    "  for obj in y_pred:\n",
    "    obj = obj[0]\n",
    "    if obj[0] == 'Non-Offensive':\n",
    "      y_pred_aux.append([obj[1], 1 - obj[1]])\n",
    "    else:\n",
    "      y_pred_aux.append([1 - obj[1], obj[1]])\n",
    "\n",
    "  y_pred = np.array(y_pred_aux, np.float32)\n",
    "\n",
    "  # Métricas:\n",
    "  res = accuracy_score(y_true_sk, y_pred_sk)\n",
    "  indice = 'SKLearn accuracy score'\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 9, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'SKLearn log loss'\n",
    "  res = log_loss(y_true_sk, y_pred_sk, eps=1e-15)\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 10, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'SKLearn precision score (micro)'\n",
    "  res = precision_score(y_true_sk, y_pred_sk, average='micro')\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 11, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'SKLearn precision score (macro)'\n",
    "  res = precision_score(y_true_sk, y_pred_sk, average='macro')\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 12, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'SKLearn recall score (micro)'\n",
    "  res = recall_score(y_true_sk, y_pred_sk, average='micro')\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 13, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'SKLearn recall score (macro)'\n",
    "  res = recall_score(y_true_sk, y_pred_sk, average='macro')\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 14, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'TFAddons recall (micro)'\n",
    "  metric = tf.keras.metrics.Recall()\n",
    "  metric.update_state(y_true, y_pred)\n",
    "  result = metric.result()\n",
    "  res = result.numpy()\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 15, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'TFAddons F1 Score (categorias)'\n",
    "  metric = tfa.metrics.F1Score(num_classes=2, threshold=0.5)\n",
    "  metric.update_state(y_true, y_pred)\n",
    "  result = metric.result()\n",
    "  res = result.numpy()\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 16, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'TFAddons F1 Score (micro)'\n",
    "  metric = tfa.metrics.F1Score(num_classes=2, threshold=0.5, average='micro')\n",
    "  metric.update_state(y_true, y_pred)\n",
    "  result = metric.result()\n",
    "  res = result.numpy()\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 17, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'TFAddons F1 Score (macro)'\n",
    "  metric = tfa.metrics.F1Score(num_classes=2, threshold=0.5, average='macro')\n",
    "  metric.update_state(y_true, y_pred)\n",
    "  result = metric.result()\n",
    "  res = result.numpy()\n",
    "  backup[indice] = str(res)\n",
    "  worksheet.write(i, 18, str(res))\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  indice = 'SKLearn classification report'\n",
    "  res = classification_report(y_true_sk, y_pred_sk, target_names=['Non-Offensive', 'Offensive'])\n",
    "  backup[indice] = res\n",
    "  worksheet.write(i, 19, res)\n",
    "  markdown_modelo += etiquetaMarkdown(str(res))\n",
    "  print(indice)\n",
    "  print(res)\n",
    "\n",
    "  print(backup)\n",
    "\n",
    "  with open(workdir+model_name+\"-evaluacion.json\", \"w\") as f:\n",
    "    json.dump(backup, f)\n",
    "\n",
    "  markdown_final += markdown_modelo + '\\n'\n",
    "  i += 1\n",
    "\n",
    "workbook.close()\n",
    "with open(workdir+'README.md', 'w') as f:\n",
    "    f.write(markdown_final)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "UyhtNxuK3E88"
   },
   "source": [
    "## Para generar gráficas de aprendizaje (antes se ha debido copiar la salida de la celda de entrenamiento en el fichero LogEntrenamiento.txt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "eZczbKxj3Imr",
    "outputId": "85ca02aa-b7a8-43f6-ce2f-c47ef7c5188b",
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[{'model': 'average_word_vec', 'lang': 'en', 'accuracy': [0.4955, 0.567, 0.6012, 0.6235, 0.6533, 0.6905, 0.7188, 0.7277, 0.7753, 0.8006, 0.7917, 0.7887, 0.8065, 0.8571, 0.8527, 0.8571, 0.875, 0.8839, 0.881, 0.8914, 0.9092, 0.8943, 0.9122, 0.9167, 0.9077, 0.9137, 0.9182, 0.9196, 0.9182, 0.9241, 0.9345, 0.9271, 0.933, 0.933, 0.9435, 0.933, 0.9449, 0.9539, 0.9539, 0.9494, 0.9658, 0.9479, 0.9554, 0.9673, 0.9494, 0.9628, 0.9658, 0.9554, 0.9673, 0.9628, 0.9673, 0.9792, 0.9747, 0.9807, 0.9821, 0.9836, 0.9717, 0.9792, 0.9777, 0.9836, 0.9851, 0.9866, 0.9881, 0.9881, 0.9821, 0.9881, 0.994, 0.9851, 0.9911, 0.9881, 0.9896, 0.9881, 0.9911, 0.9896, 0.9911, 0.994, 1.0, 0.9955, 0.994, 0.9896, 0.9955, 0.994, 0.9955, 0.997, 0.997, 0.9955, 0.994, 0.997, 0.997, 0.9955, 0.9955, 0.9985, 0.997, 0.9985, 0.997, 0.997, 0.9985, 1.0, 0.997, 1.0, 1.0, 0.997, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 0.9985, 1.0, 0.9955, 1.0, 1.0, 1.0, 0.9985, 0.9985, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 0.9985, 0.997, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 0.9985, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.9985, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], 'loss': [0.6932, 0.6922, 0.6906, 0.6893, 0.6869, 0.6842, 0.6819, 0.6795, 0.6745, 0.6699, 0.6636, 0.6574, 0.6544, 0.6429, 0.6326, 0.6252, 0.6151, 0.6029, 0.5913, 0.5827, 0.5602, 0.5492, 0.5312, 0.5193, 0.5051, 0.4902, 0.4689, 0.4535, 0.4402, 0.4189, 0.4024, 0.3884, 0.3734, 0.3596, 0.3412, 0.3228, 0.3109, 0.2943, 0.2802, 0.2688, 0.2452, 0.2424, 0.2272, 0.2107, 0.2086, 0.1978, 0.1876, 0.1827, 0.1654, 0.1636, 0.1465, 0.1327, 0.1356, 0.128, 0.111, 0.1199, 0.116, 0.1051, 0.1, 0.0984, 0.0944, 0.0881, 0.0807, 0.072, 0.0773, 0.0752, 0.0627, 0.0663, 0.0579, 0.0582, 0.0525, 0.0504, 0.047, 0.0503, 0.0478, 0.0366, 0.0343, 0.034, 0.0394, 0.0378, 0.0327, 0.0313, 0.0284, 0.0271, 0.0307, 0.0264, 0.0303, 0.0251, 0.0199, 0.0255, 0.0228, 0.0212, 0.017, 0.0163, 0.0166, 0.0152, 0.0176, 0.0128, 0.0138, 0.0136, 0.0109, 0.0131, 0.0107, 0.0085, 0.0102, 0.0106, 0.007, 0.0092, 0.0103, 0.0064, 0.0104, 0.0075, 0.0076, 0.0078, 0.0105, 0.0053, 0.007, 0.0047, 0.006, 0.0056, 0.0066, 0.0026, 0.0053, 0.0056, 0.0063, 0.0048, 0.0039, 0.0038, 0.0058, 0.0054, 0.0038, 0.0027, 0.0026, 0.0027, 0.0039, 0.0028, 0.0059, 0.0044, 0.0031, 0.0038, 0.0042, 0.0033, 0.0025, 0.0026, 0.0037, 0.0011, 0.0024, 0.0021, 0.0024, 0.0025, 0.0014, 0.0024, 0.0029, 0.0026, 0.0022, 0.00078449, 0.0015, 0.0017, 0.0012, 0.0012, 0.0026, 0.0021, 0.0019, 0.0015, 0.0016, 0.00077171, 0.0011, 0.0012, 0.0016, 0.0006238, 0.00053922, 0.0014, 0.00094344, 0.0019, 0.00067529, 0.00083292, 0.0026, 0.0017, 0.0019, 0.001, 0.00044881, 0.0015, 0.0014, 0.00061012, 0.00048455, 0.0003933, 0.0011, 0.0011, 0.00068837, 0.0013, 0.0015, 0.0019, 0.0012, 0.00040702, 0.0013, 0.00038033, 0.0006159, 0.00082647, 0.00042215, 0.0024, 0.00063931, 0.00045687, 0.00016821, 0.00076737, 0.00068828, 0.0023, 0.00026204, 0.00016244, 0.00025107, 0.00093042, 0.00090069, 0.00020344, 0.00071527, 0.00047694, 0.0001286, 0.00053977, 0.00044533, 0.0001215, 0.00039569, 0.0015, 0.00080992, 0.00015306, 0.00076684, 0.0015, 0.00032466, 0.00011183, 0.00033468, 0.00044397, 0.00027127, 0.0043, 0.00063901, 0.00027526, 0.00022735, 0.001, 0.00036294, 0.00073835, 0.0015, 0.0001105, 8.7335e-05, 9.3622e-05, 0.00057042, 0.00010543, 0.00039886, 0.00040255, 0.00047686, 0.00022066, 0.00018501, 0.0006462, 0.00035121, 6.064e-05, 0.00032554, 5.1833e-05, 0.0015, 0.00017319, 0.00051789, 0.0018, 0.0001933, 0.00033293, 0.00015399, 5.6352e-05, 6.877e-05, 0.00080985, 0.00029403, 0.00052386, 0.00048223, 0.00075593, 6.87e-05, 5.6166e-05, 0.00010816, 0.00028258, 9.9918e-05, 7.1322e-05, 6.4368e-05, 0.00030142, 0.0011, 0.0014, 0.00041496, 0.00038437, 0.00013748, 7.1278e-05, 5.5471e-05, 0.00051824, 0.00015574, 0.00029899, 4.7663e-05, 0.00044625, 1.3161e-05, 3.4204e-05, 8.3613e-05, 1.5377e-05, 5.4995e-05, 0.00026224, 0.00028864, 0.00031062, 0.00067111, 0.00010021, 3.4161e-05, 0.00046227, 5.8682e-05, 0.00056392]}, {'model': 'mobilebert', 'lang': 'en', 'accuracy': [0.7068, 0.942, 0.9702], 'loss': [1.7974, 0.1989, 0.0874]}]\n"
     ]
    }
   ],
   "source": [
    "# Cargamos las etiquetas\n",
    "backup = [\n",
    "          {'model': \"average_word_vec\", 'lang': 'en', 'accuracy': [], 'loss': []},\n",
    "          {'model': \"mobilebert\", 'lang': 'en', 'accuracy': [], 'loss': []}\n",
    "         ]\n",
    "i = -1\n",
    "skipNextLine = False\n",
    "with open(workdir+'LogEntrenamiento.txt','r') as f:\n",
    "    for line in f:\n",
    "      if skipNextLine:\n",
    "        skipNextLine = False\n",
    "        continue\n",
    "      if \"Entrenando modelo \" in line:\n",
    "        i += 1\n",
    "        #print(line)\n",
    "      elif \"Calculando accuracy y loss final:\" in line:\n",
    "        skipNextLine = True\n",
    "      elif \" - loss: \" in line:\n",
    "        l = line.strip().split(' ')\n",
    "        loss = float(l[7])\n",
    "        accuracy = float(l[10])\n",
    "        backup[i]['accuracy'].append(accuracy)\n",
    "        backup[i]['loss'].append(loss)\n",
    "print(backup)\n",
    "\n",
    "# Lo guardamos en un JSON\n",
    "with open(workdir+'graficas_entrenamiento.json','w') as f:\n",
    "  json.dump(backup, f, indent=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generamos las gráficas de aprendizaje de los distintos modelos durante su entrenamiento"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "c3vTVsWW7mKz",
    "outputId": "202330b3-78e7-4a23-bc7f-9d8a3afb188f"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "for model in backup:\n",
    "  x  = range(1, len(model['accuracy'])+1)\n",
    "  y1 = model['accuracy']\n",
    "  y2 = model['loss']\n",
    "  plt.plot(x, y1, label=\"accuracy\")\n",
    "  plt.plot(x, y2, label=\"loss\")\n",
    "  plt.plot()\n",
    "\n",
    "  plt.xlabel(\"Epochs\")\n",
    "  plt.ylabel(\"Learning evolution\")\n",
    "  plt.title(\"Learning evolution of \"+model['model']+\" - \"+model['lang'])\n",
    "  plt.legend()\n",
    "  plt.show()"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "provenance": [],
   "toc_visible": true
  },
  "gpuClass": "standard",
  "kernelspec": {
   "display_name": "Python 3.9",
   "language": "python",
   "name": "python3.9"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}