003_train_LSTM.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "21bee0a3-79a7-43b2-9ac2-15cad5630e22",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "353a2d75-f0f4-4563-9842-c68d77118161",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv('data/acetylcholinesterase_02_bioactivity_data_preprocessed.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d1554fa5-d9ee-456c-be20-80571e319fc6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>molecule_chembl_id</th>\n",
       "      <th>canonical_smiles</th>\n",
       "      <th>standard_value</th>\n",
       "      <th>standard_value_norm</th>\n",
       "      <th>pIC50</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>CHEMBL133897</td>\n",
       "      <td>CCOc1nn(-c2cccc(OCc3ccccc3)c2)c(=O)o1</td>\n",
       "      <td>750.0</td>\n",
       "      <td>750.0</td>\n",
       "      <td>6.124939</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>CHEMBL336398</td>\n",
       "      <td>O=C(N1CCCCC1)n1nc(-c2ccc(Cl)cc2)nc1SCC1CC1</td>\n",
       "      <td>100.0</td>\n",
       "      <td>100.0</td>\n",
       "      <td>7.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>CHEMBL131588</td>\n",
       "      <td>CN(C(=O)n1nc(-c2ccc(Cl)cc2)nc1SCC(F)(F)F)c1ccccc1</td>\n",
       "      <td>50000.0</td>\n",
       "      <td>50000.0</td>\n",
       "      <td>4.301030</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>CHEMBL130628</td>\n",
       "      <td>O=C(N1CCCCC1)n1nc(-c2ccc(Cl)cc2)nc1SCC(F)(F)F</td>\n",
       "      <td>300.0</td>\n",
       "      <td>300.0</td>\n",
       "      <td>6.522879</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CHEMBL130478</td>\n",
       "      <td>CSc1nc(-c2ccc(OC(F)(F)F)cc2)nn1C(=O)N(C)C</td>\n",
       "      <td>800.0</td>\n",
       "      <td>800.0</td>\n",
       "      <td>6.096910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5038</th>\n",
       "      <td>CHEMBL4554172</td>\n",
       "      <td>Cc1ccc(-c2nc(NC(=O)C3CCN(Cc4cccc(F)c4)CC3)sc2-...</td>\n",
       "      <td>10000.0</td>\n",
       "      <td>10000.0</td>\n",
       "      <td>5.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5039</th>\n",
       "      <td>CHEMBL4533844</td>\n",
       "      <td>Cc1ccc(-c2nc(NC(=O)C3CCN(Cc4ccccc4C(F)(F)F)CC3...</td>\n",
       "      <td>7570.0</td>\n",
       "      <td>7570.0</td>\n",
       "      <td>5.120904</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5040</th>\n",
       "      <td>CHEMBL4570655</td>\n",
       "      <td>Cc1ccc(-c2nc(NC(=O)C3CCN(Cc4ccccc4C)CC3)sc2-c2...</td>\n",
       "      <td>10000.0</td>\n",
       "      <td>10000.0</td>\n",
       "      <td>5.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5041</th>\n",
       "      <td>CHEMBL4571704</td>\n",
       "      <td>COc1cc(O)c2c(c1)C1=C(c3c(O)c(O)cc(C)c3-c3cc(OC...</td>\n",
       "      <td>15500.0</td>\n",
       "      <td>15500.0</td>\n",
       "      <td>4.809668</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5042</th>\n",
       "      <td>CHEMBL4556664</td>\n",
       "      <td>COc1cc(O)c(C(C)=O)c(-c2c(C)cc(O)c3oc4c(O)cc(C)...</td>\n",
       "      <td>68300.0</td>\n",
       "      <td>68300.0</td>\n",
       "      <td>4.165579</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5043 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     molecule_chembl_id                                   canonical_smiles  \\\n",
       "0          CHEMBL133897              CCOc1nn(-c2cccc(OCc3ccccc3)c2)c(=O)o1   \n",
       "1          CHEMBL336398         O=C(N1CCCCC1)n1nc(-c2ccc(Cl)cc2)nc1SCC1CC1   \n",
       "2          CHEMBL131588  CN(C(=O)n1nc(-c2ccc(Cl)cc2)nc1SCC(F)(F)F)c1ccccc1   \n",
       "3          CHEMBL130628      O=C(N1CCCCC1)n1nc(-c2ccc(Cl)cc2)nc1SCC(F)(F)F   \n",
       "4          CHEMBL130478          CSc1nc(-c2ccc(OC(F)(F)F)cc2)nn1C(=O)N(C)C   \n",
       "...                 ...                                                ...   \n",
       "5038      CHEMBL4554172  Cc1ccc(-c2nc(NC(=O)C3CCN(Cc4cccc(F)c4)CC3)sc2-...   \n",
       "5039      CHEMBL4533844  Cc1ccc(-c2nc(NC(=O)C3CCN(Cc4ccccc4C(F)(F)F)CC3...   \n",
       "5040      CHEMBL4570655  Cc1ccc(-c2nc(NC(=O)C3CCN(Cc4ccccc4C)CC3)sc2-c2...   \n",
       "5041      CHEMBL4571704  COc1cc(O)c2c(c1)C1=C(c3c(O)c(O)cc(C)c3-c3cc(OC...   \n",
       "5042      CHEMBL4556664  COc1cc(O)c(C(C)=O)c(-c2c(C)cc(O)c3oc4c(O)cc(C)...   \n",
       "\n",
       "      standard_value  standard_value_norm     pIC50  \n",
       "0              750.0                750.0  6.124939  \n",
       "1              100.0                100.0  7.000000  \n",
       "2            50000.0              50000.0  4.301030  \n",
       "3              300.0                300.0  6.522879  \n",
       "4              800.0                800.0  6.096910  \n",
       "...              ...                  ...       ...  \n",
       "5038         10000.0              10000.0  5.000000  \n",
       "5039          7570.0               7570.0  5.120904  \n",
       "5040         10000.0              10000.0  5.000000  \n",
       "5041         15500.0              15500.0  4.809668  \n",
       "5042         68300.0              68300.0  4.165579  \n",
       "\n",
       "[5043 rows x 5 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2985df76-23b9-45a8-83ed-ce8ae510219f",
   "metadata": {},
   "source": [
    "# Canonical_smiles To secuence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "87e524d2-e7e9-4d3b-b1f1-b62254a0bc66",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.preprocessing.text import Tokenizer\n",
    "from tensorflow.keras.preprocessing.sequence import pad_sequences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "07f8fff5-c1f6-48d6-9e19-ee5defde20b1",
   "metadata": {},
   "outputs": [],
   "source": [
    "max_len_idx = df['canonical_smiles'].apply(len).argmax()\n",
    "min_len_idx = df['canonical_smiles'].apply(len).argmin()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "703325ab-0193-4bf5-9b66-a880d7bc346b",
   "metadata": {},
   "outputs": [],
   "source": [
    "X = df['canonical_smiles']\n",
    "y = df['pIC50']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "cc67f16c-b0b6-4aee-8faa-559d9cd1abab",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'c': 1,\n",
       " 'C': 2,\n",
       " '(': 3,\n",
       " ')': 4,\n",
       " '1': 5,\n",
       " 'O': 6,\n",
       " '2': 7,\n",
       " 'N': 8,\n",
       " '=': 9,\n",
       " '3': 10,\n",
       " '[': 11,\n",
       " ']': 12,\n",
       " '@': 13,\n",
       " 'n': 14,\n",
       " '4': 15,\n",
       " 'H': 16,\n",
       " '-': 17,\n",
       " '/': 18,\n",
       " 'l': 19,\n",
       " '.': 20,\n",
       " '+': 21,\n",
       " 'F': 22,\n",
       " '5': 23,\n",
       " 'o': 24,\n",
       " 'B': 25,\n",
       " 'r': 26,\n",
       " 'S': 27,\n",
       " '\\\\': 28,\n",
       " '#': 29,\n",
       " '6': 30,\n",
       " 's': 31,\n",
       " 'I': 32,\n",
       " 'e': 33,\n",
       " 'P': 34,\n",
       " '7': 35,\n",
       " '8': 36,\n",
       " 'i': 37,\n",
       " 'a': 38}"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tk = Tokenizer(char_level=True, lower=False)\n",
    "tk.fit_on_texts(X)\n",
    "tk.word_index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "7c5e1393-67f1-4aa0-ba7d-d27eb7081a02",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_seq = tk.texts_to_sequences(X)\n",
    "X_seq_pad = pad_sequences(X_seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "a9ecefbc-828d-488b-b456-6e746e1cbc0e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 2, 11,  2, 13, 13, 12,  5,  7,  2,  2, 11,  2, 13, 16, 12, 10, 11,\n",
       "        2, 13, 12, 15,  3,  2,  4,  2,  2,  2, 11,  2, 13, 13, 12, 23,  3,\n",
       "        2,  4,  2,  3,  9,  6,  4,  6,  2,  2, 11,  8, 21, 12,  3,  2,  4,\n",
       "        3,  2,  4,  2,  2,  2,  2,  2,  2,  2,  2,  2, 11,  8, 21, 12,  3,\n",
       "        2,  4,  3,  2,  4,  2,  2,  6,  2,  3,  9,  6,  4, 11,  2, 13, 12,\n",
       "       30,  3,  2,  4,  2,  2,  2, 11,  2, 13, 12, 35,  3,  2,  4, 11,  2,\n",
       "       13, 13, 16, 12, 30,  2,  2, 11,  2, 13, 13, 12, 30, 36,  2, 11,  2,\n",
       "       13, 13, 16, 12,  3,  6,  2,  3,  9,  6,  4,  2,  2,  2,  2,  2,  2,\n",
       "        2,  2,  2,  3,  9,  6,  4,  6, 11,  2, 13, 13, 16, 12,  5,  2, 11,\n",
       "        2, 13, 13, 12, 10,  3,  2,  2, 11,  2, 13, 13, 16, 12, 15, 23,  4,\n",
       "        2,  7,  4, 11,  2, 13, 13, 12,  3,  2,  4,  3,  2,  2, 11,  2, 13,\n",
       "       13, 16, 12, 35, 30,  4,  2, 36, 20, 11, 25, 26, 17, 12, 20, 11, 25,\n",
       "       26, 17, 12], dtype=int32)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_seq_pad[max_len_idx]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "14d2a648-a9dc-482e-b3aa-4a56bd3b8437",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 8, 2, 3, 9, 6, 4, 6], dtype=int32)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_seq_pad[min_len_idx]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4233710d-9274-4dae-b22d-cf42b57ad862",
   "metadata": {},
   "source": [
    "# Split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "ce7a95e6-2486-42fb-97fb-865e3fd6a561",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "fcb9840f-2f41-45be-99a5-f05f3ab73ed8",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X_seq_pad, y, test_size=0.2, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "4ad17134-3e15-42d1-b8c9-0c793169e145",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(4034, 4034, 1009)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(X_train), len(y_train), len(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "547d87a7-98e3-442e-9f7b-2c4d9d6600de",
   "metadata": {},
   "source": [
    "# Data augmentation (Optional)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 630,
   "id": "5ddd8741-95b9-4853-b5af-a7ad0d9f92c2",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train = np.vstack([X_train, X_train[:, ::-1]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 631,
   "id": "40594f07-9dee-4b22-8e67-99783778485e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8068, 207)"
      ]
     },
     "execution_count": 631,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 633,
   "id": "fda1f529-e7f0-41d1-b3e8-9f8fc49181b1",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train = np.hstack([y_train, y_train])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 634,
   "id": "72d7302f-351b-45cc-974c-35288a31d556",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8068,)"
      ]
     },
     "execution_count": 634,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "85941430-2534-4d7d-ba52-d934a31c1bf6",
   "metadata": {},
   "source": [
    "# Network Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "6bd5d1bf-4640-4cbc-acd3-72881ddcc07e",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras import Sequential\n",
    "from tensorflow.keras.layers import LSTM, Dense, Embedding, Bidirectional, Dropout, Activation, BatchNormalization\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint\n",
    "from tensorflow.keras.optimizers import Adam, RMSprop\n",
    "from tensorflow.keras import backend as K"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "4bf115a2-9f1b-42b8-9b62-59728f1da977",
   "metadata": {},
   "outputs": [],
   "source": [
    "def R2(y_true, y_pred):\n",
    "    SS_res =  K.sum(K.square( y_true-y_pred ))\n",
    "    SS_tot = K.sum(K.square( y_true - K.mean(y_true) ) )\n",
    "    return ( 1 - SS_res/(SS_tot + K.epsilon()) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "92d30b68-3110-459c-887d-a06f056d067b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "38"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocab_size = len(tk.word_index)\n",
    "vocab_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "6166a72b-ccd1-459c-a3a1-fa43bf31a369",
   "metadata": {},
   "outputs": [],
   "source": [
    "max_sequence_len = X_seq_pad.shape[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "475ce476-98ae-4459-bdbf-145fdee398d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Embedding(vocab_size + 1, 64, input_length=max_sequence_len))\n",
    "model.add(Bidirectional(LSTM(100, return_sequences=False, activation='tanh')))\n",
    "model.add(Dense(50))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Activation('relu'))\n",
    "model.add(Dense(1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "237c1327-028f-4f70-bd29-ea94cd999686",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding (Embedding)        (None, 207, 64)           2496      \n",
      "_________________________________________________________________\n",
      "bidirectional (Bidirectional (None, 200)               132000    \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 50)                10050     \n",
      "_________________________________________________________________\n",
      "batch_normalization (BatchNo (None, 50)                200       \n",
      "_________________________________________________________________\n",
      "activation (Activation)      (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 1)                 51        \n",
      "=================================================================\n",
      "Total params: 144,797\n",
      "Trainable params: 144,697\n",
      "Non-trainable params: 100\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "9138b411-fc5d-41bd-8c8e-c12580cc5b14",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer=RMSprop(learning_rate=0.001), loss='mse', metrics=[R2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "62c3f792-f413-4150-8ebf-1f59c5fb986d",
   "metadata": {},
   "outputs": [],
   "source": [
    "mcp = ModelCheckpoint('models/best_model_{epoch}', save_best_only=True, save_format=\"h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "3f37ab4e-0e3a-44c6-a572-ba5c67930c61",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "32/32 [==============================] - 8s 183ms/step - loss: 19.9009 - R2: -6.5828 - val_loss: 26.6316 - val_R2: -8.0484\n",
      "Epoch 2/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 11.4554 - R2: -3.4102 - val_loss: 19.5925 - val_R2: -5.6488\n",
      "Epoch 3/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 7.3339 - R2: -1.8244 - val_loss: 14.3877 - val_R2: -3.8734\n",
      "Epoch 4/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 4.4449 - R2: -0.6587 - val_loss: 9.1465 - val_R2: -2.0879\n",
      "Epoch 5/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 2.8682 - R2: -0.0578 - val_loss: 7.6999 - val_R2: -1.5981\n",
      "Epoch 6/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 2.4432 - R2: 0.1067 - val_loss: 11.7625 - val_R2: -2.9851\n",
      "Epoch 7/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 2.2959 - R2: 0.1685 - val_loss: 7.4159 - val_R2: -1.5045\n",
      "Epoch 8/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 2.1774 - R2: 0.1934 - val_loss: 3.9992 - val_R2: -0.3372\n",
      "Epoch 9/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 2.1302 - R2: 0.2221 - val_loss: 4.3937 - val_R2: -0.4690\n",
      "Epoch 10/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 2.0777 - R2: 0.2292 - val_loss: 4.2235 - val_R2: -0.4101\n",
      "Epoch 11/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 1.9955 - R2: 0.2635 - val_loss: 2.4704 - val_R2: 0.1783\n",
      "Epoch 12/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.9248 - R2: 0.2925 - val_loss: 2.9731 - val_R2: 0.0030\n",
      "Epoch 13/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.8264 - R2: 0.3268 - val_loss: 2.1745 - val_R2: 0.2711\n",
      "Epoch 14/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.7912 - R2: 0.3357 - val_loss: 5.6287 - val_R2: -0.8954\n",
      "Epoch 15/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.7726 - R2: 0.3498 - val_loss: 2.7550 - val_R2: 0.0800\n",
      "Epoch 16/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 1.7161 - R2: 0.3692 - val_loss: 2.4811 - val_R2: 0.1756\n",
      "Epoch 17/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.7089 - R2: 0.3689 - val_loss: 20.5612 - val_R2: -6.0594\n",
      "Epoch 18/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 1.6759 - R2: 0.3765 - val_loss: 4.6172 - val_R2: -0.5592\n",
      "Epoch 19/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.6354 - R2: 0.3843 - val_loss: 1.9962 - val_R2: 0.3357\n",
      "Epoch 20/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 1.5897 - R2: 0.4045 - val_loss: 2.9126 - val_R2: 0.0158\n",
      "Epoch 21/100\n",
      "32/32 [==============================] - 5s 169ms/step - loss: 1.5696 - R2: 0.4104 - val_loss: 25.8186 - val_R2: -7.8123\n",
      "Epoch 22/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.5545 - R2: 0.4200 - val_loss: 9.7508 - val_R2: -2.3223\n",
      "Epoch 23/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.8083 - R2: 0.3240 - val_loss: 3.3207 - val_R2: -0.1233\n",
      "Epoch 24/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.4991 - R2: 0.4459 - val_loss: 7.3732 - val_R2: -1.4974\n",
      "Epoch 25/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.4646 - R2: 0.4498 - val_loss: 1.9633 - val_R2: 0.3351\n",
      "Epoch 26/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.4431 - R2: 0.4604 - val_loss: 2.6559 - val_R2: 0.1098\n",
      "Epoch 27/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.4824 - R2: 0.4464 - val_loss: 5.8978 - val_R2: -0.9988\n",
      "Epoch 28/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.3376 - R2: 0.5040 - val_loss: 5.3350 - val_R2: -0.8102\n",
      "Epoch 29/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.3660 - R2: 0.4846 - val_loss: 3.3506 - val_R2: -0.1318\n",
      "Epoch 30/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.3733 - R2: 0.4918 - val_loss: 2.1224 - val_R2: 0.2861\n",
      "Epoch 31/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.3199 - R2: 0.5052 - val_loss: 2.1624 - val_R2: 0.2708\n",
      "Epoch 32/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.2826 - R2: 0.5180 - val_loss: 1.7374 - val_R2: 0.4151\n",
      "Epoch 33/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.2673 - R2: 0.5248 - val_loss: 2.7720 - val_R2: 0.0653\n",
      "Epoch 34/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.2833 - R2: 0.5266 - val_loss: 2.0232 - val_R2: 0.3153\n",
      "Epoch 35/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.2543 - R2: 0.5324 - val_loss: 1.9404 - val_R2: 0.3467\n",
      "Epoch 36/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.2183 - R2: 0.5449 - val_loss: 1.9052 - val_R2: 0.3513\n",
      "Epoch 37/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.1869 - R2: 0.5514 - val_loss: 2.0833 - val_R2: 0.2938\n",
      "Epoch 38/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.1740 - R2: 0.5609 - val_loss: 3.2735 - val_R2: -0.1235\n",
      "Epoch 39/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.1521 - R2: 0.5395 - val_loss: 1.9312 - val_R2: 0.3484\n",
      "Epoch 40/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.1248 - R2: 0.5823 - val_loss: 1.5241 - val_R2: 0.4858\n",
      "Epoch 41/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.1051 - R2: 0.5828 - val_loss: 6.0889 - val_R2: -1.0518\n",
      "Epoch 42/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.0765 - R2: 0.5969 - val_loss: 5.1596 - val_R2: -0.7369\n",
      "Epoch 43/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 1.0539 - R2: 0.6037 - val_loss: 4.2484 - val_R2: -0.4298\n",
      "Epoch 44/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 1.0628 - R2: 0.6020 - val_loss: 1.8506 - val_R2: 0.3801\n",
      "Epoch 45/100\n",
      "32/32 [==============================] - 5s 171ms/step - loss: 1.0561 - R2: 0.6052 - val_loss: 5.2614 - val_R2: -0.7596\n",
      "Epoch 46/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.9832 - R2: 0.6293 - val_loss: 4.8585 - val_R2: -0.6222\n",
      "Epoch 47/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.9619 - R2: 0.6374 - val_loss: 3.4265 - val_R2: -0.1556\n",
      "Epoch 48/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.9828 - R2: 0.6311 - val_loss: 2.2010 - val_R2: 0.2567\n",
      "Epoch 49/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.9369 - R2: 0.6436 - val_loss: 2.1500 - val_R2: 0.2697\n",
      "Epoch 50/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.9537 - R2: 0.6436 - val_loss: 2.1080 - val_R2: 0.2842\n",
      "Epoch 51/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.9262 - R2: 0.6554 - val_loss: 2.3300 - val_R2: 0.2061\n",
      "Epoch 52/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.8819 - R2: 0.6668 - val_loss: 2.8542 - val_R2: 0.0320\n",
      "Epoch 53/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.8570 - R2: 0.6730 - val_loss: 2.9162 - val_R2: 0.0119\n",
      "Epoch 54/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.8283 - R2: 0.6894 - val_loss: 2.5825 - val_R2: 0.1257\n",
      "Epoch 55/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.8520 - R2: 0.6789 - val_loss: 4.0363 - val_R2: -0.3742\n",
      "Epoch 56/100\n",
      "32/32 [==============================] - 5s 171ms/step - loss: 0.8332 - R2: 0.6899 - val_loss: 1.4682 - val_R2: 0.4952\n",
      "Epoch 57/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.8228 - R2: 0.6947 - val_loss: 2.0527 - val_R2: 0.3026\n",
      "Epoch 58/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.7748 - R2: 0.7060 - val_loss: 1.4803 - val_R2: 0.4945\n",
      "Epoch 59/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.7773 - R2: 0.7091 - val_loss: 2.4172 - val_R2: 0.1788\n",
      "Epoch 60/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.7486 - R2: 0.7191 - val_loss: 1.3796 - val_R2: 0.5303\n",
      "Epoch 61/100\n",
      "32/32 [==============================] - 5s 171ms/step - loss: 0.7468 - R2: 0.7214 - val_loss: 1.4596 - val_R2: 0.5000\n",
      "Epoch 62/100\n",
      "32/32 [==============================] - 6s 173ms/step - loss: 0.7035 - R2: 0.7381 - val_loss: 1.4093 - val_R2: 0.5165\n",
      "Epoch 63/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.7002 - R2: 0.7387 - val_loss: 2.0405 - val_R2: 0.2950\n",
      "Epoch 64/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.7048 - R2: 0.7316 - val_loss: 2.2542 - val_R2: 0.2353\n",
      "Epoch 65/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.6901 - R2: 0.7403 - val_loss: 2.0461 - val_R2: 0.3090\n",
      "Epoch 66/100\n",
      "32/32 [==============================] - 5s 170ms/step - loss: 0.6642 - R2: 0.7503 - val_loss: 1.2632 - val_R2: 0.5663\n",
      "Epoch 67/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.6430 - R2: 0.7605 - val_loss: 3.4611 - val_R2: -0.1782\n",
      "Epoch 68/100\n",
      "32/32 [==============================] - 6s 176ms/step - loss: 0.6532 - R2: 0.7555 - val_loss: 1.2906 - val_R2: 0.5525\n",
      "Epoch 69/100\n",
      "32/32 [==============================] - 6s 173ms/step - loss: 0.6019 - R2: 0.7755 - val_loss: 2.2570 - val_R2: 0.2256\n",
      "Epoch 70/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.6368 - R2: 0.7604 - val_loss: 2.1703 - val_R2: 0.2600\n",
      "Epoch 71/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.6050 - R2: 0.7718 - val_loss: 3.5657 - val_R2: -0.2146\n",
      "Epoch 72/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5730 - R2: 0.7872 - val_loss: 2.4019 - val_R2: 0.1821\n",
      "Epoch 73/100\n",
      "32/32 [==============================] - 6s 172ms/step - loss: 0.5836 - R2: 0.7801 - val_loss: 1.2595 - val_R2: 0.5678\n",
      "Epoch 74/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5569 - R2: 0.7929 - val_loss: 2.0334 - val_R2: 0.3031\n",
      "Epoch 75/100\n",
      "32/32 [==============================] - 6s 172ms/step - loss: 0.5504 - R2: 0.7956 - val_loss: 1.3015 - val_R2: 0.5517\n",
      "Epoch 76/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5623 - R2: 0.7892 - val_loss: 2.9496 - val_R2: -0.0043\n",
      "Epoch 77/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5367 - R2: 0.7950 - val_loss: 4.7585 - val_R2: -0.6191\n",
      "Epoch 78/100\n",
      "32/32 [==============================] - 6s 174ms/step - loss: 0.5159 - R2: 0.8110 - val_loss: 3.0846 - val_R2: -0.0422\n",
      "Epoch 79/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5113 - R2: 0.8140 - val_loss: 2.1276 - val_R2: 0.2755\n",
      "Epoch 80/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5020 - R2: 0.8064 - val_loss: 1.2673 - val_R2: 0.5608\n",
      "Epoch 81/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5024 - R2: 0.8091 - val_loss: 1.4620 - val_R2: 0.4998\n",
      "Epoch 82/100\n",
      "32/32 [==============================] - 6s 173ms/step - loss: 0.5015 - R2: 0.8136 - val_loss: 1.4548 - val_R2: 0.4981\n",
      "Epoch 83/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4724 - R2: 0.8236 - val_loss: 1.6242 - val_R2: 0.4415\n",
      "Epoch 84/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4680 - R2: 0.8251 - val_loss: 1.2469 - val_R2: 0.5710\n",
      "Epoch 85/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.5029 - R2: 0.8156 - val_loss: 1.2091 - val_R2: 0.5784\n",
      "Epoch 86/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4702 - R2: 0.8225 - val_loss: 1.6259 - val_R2: 0.4448\n",
      "Epoch 87/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4437 - R2: 0.8367 - val_loss: 1.1977 - val_R2: 0.5874\n",
      "Epoch 88/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4398 - R2: 0.8406 - val_loss: 1.5190 - val_R2: 0.4793\n",
      "Epoch 89/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4575 - R2: 0.8303 - val_loss: 1.2181 - val_R2: 0.5795\n",
      "Epoch 90/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4426 - R2: 0.8354 - val_loss: 1.4078 - val_R2: 0.5163\n",
      "Epoch 91/100\n",
      "32/32 [==============================] - 6s 173ms/step - loss: 0.4349 - R2: 0.8349 - val_loss: 1.8217 - val_R2: 0.3746\n",
      "Epoch 92/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4234 - R2: 0.8396 - val_loss: 1.5823 - val_R2: 0.4553\n",
      "Epoch 93/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4245 - R2: 0.8436 - val_loss: 1.8075 - val_R2: 0.3833\n",
      "Epoch 94/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4022 - R2: 0.8530 - val_loss: 1.7979 - val_R2: 0.3879\n",
      "Epoch 95/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4132 - R2: 0.8489 - val_loss: 1.3050 - val_R2: 0.5525\n",
      "Epoch 96/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.4247 - R2: 0.8439 - val_loss: 2.3723 - val_R2: 0.1788\n",
      "Epoch 97/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.3887 - R2: 0.8562 - val_loss: 1.2367 - val_R2: 0.5719\n",
      "Epoch 98/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.3807 - R2: 0.8616 - val_loss: 1.2472 - val_R2: 0.5677\n",
      "Epoch 99/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.3682 - R2: 0.8658 - val_loss: 1.4606 - val_R2: 0.4938\n",
      "Epoch 100/100\n",
      "32/32 [==============================] - 5s 172ms/step - loss: 0.3732 - R2: 0.8622 - val_loss: 1.2123 - val_R2: 0.5829\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f65541d5438>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "history = model.fit(X_train, y_train, epochs=100, batch_size=128, validation_data=(X_test, y_test), callbacks=[])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "0084fbb1-e83e-4808-99bf-1458f3ea3076",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = model.predict(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "382979a3-f299-4762-8b6d-10a6ea4d3c70",
   "metadata": {},
   "source": [
    "# Score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "f9fd5d60-841f-46c3-94b6-5755b4d548cd",
   "metadata": {},
   "outputs": [],
   "source": [
    "r2 = 1 - ((y_test.values - y_pred.reshape(-1)) ** 2).sum() / ((y_test - y_test.mean()) ** 2).sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "4d2bfabe-2e3e-40cf-a1e8-2d458ee7f328",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-2.2295744744463684"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "r2\n",
    "# 0.4984533246797399"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "85262541-e04a-4ff2-b3e6-a47643db66f6",
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.save('models/embed_64_bi_lstm_100_dense_50_bn.h5', )"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f683e18-c876-4f7c-93ea-a33d495dcdaa",
   "metadata": {},
   "source": [
    "# Load model (solo si se grabo el modelo)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "285c8740-f0bb-4237-8593-ff9b815de0e3",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.models import load_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 473,
   "id": "2168984f-770c-436e-92da-c1344b1d3abc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# model_loaded = load_model('models/embed_64_bi_lstm_100_dense_50_bn.h5', custom_objects={'R2': R2})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "13a9d7d8-41e0-44a0-a8b9-02157a11f8f2",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = model_loaded.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 475,
   "id": "26c63493-3f25-4383-9654-58ee62aa3ae5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6913848492905522"
      ]
     },
     "execution_count": 475,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "r2 = 1 - ((y_test.values - y_pred.reshape(-1)) ** 2).sum() / ((y_test - y_test.mean()) ** 2).sum()\n",
    "r2"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7f1a6260-e31f-4e5e-a923-ce67bf31152f",
   "metadata": {},
   "source": [
    "# Evaluación"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "7dd9d7e6-2789-459e-b9dd-593e10847fc4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<function matplotlib.pyplot.show(close=None, block=None)>"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "sns.set(color_codes=True)\n",
    "sns.set_style(\"white\")\n",
    "\n",
    "ax = sns.regplot(x=y_test, y=y_pred, scatter_kws={'alpha':0.4})\n",
    "ax.set_xlabel('Experimental pIC50', fontsize='large', fontweight='bold')\n",
    "ax.set_ylabel('Predicted pIC50', fontsize='large', fontweight='bold')\n",
    "ax.set_xlim(0, 12)\n",
    "ax.set_ylim(0, 12)\n",
    "ax.figure.set_size_inches(5, 5)\n",
    "plt.show"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "76f567be-4a4a-4e73-998a-575a6afbf603",
   "metadata": {},
   "source": [
    "# TSNE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "77f4b582-a35b-42db-896d-e2986a77c521",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.manifold import TSNE\n",
    "from matplotlib import pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "4927e494-6cb0-4598-93ff-4efd5363b6fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "embedings = model.layers[0].get_weights()[0][1:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "1033e69d-2e6c-4a56-9237-8f8063c0b8b8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(38, 64)"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embedings.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "321d751f-a071-40ee-995e-6b111090a782",
   "metadata": {},
   "outputs": [],
   "source": [
    "tsne = TSNE(2)\n",
    "X_2d = tsne.fit_transform(embedings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "922a0be0-332b-4f08-9455-77b02e1c33bf",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYQAAAD7CAYAAABqvuNzAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8rg+JYAAAACXBIWXMAAAsTAAALEwEAmpwYAAAmJUlEQVR4nO3de1hU1eI+8BdGB0tUwABH8YqpKKIoeOFRCZAgwhtWmoqpWd5LC4wsFSgzb8eOHi9xUjQ1Nb8pKhheqB/g0Qw0C9PUSDMBQS6Kigw4M78/cibwAoozs/bMvJ/n6Qlm+8ALjvPOXnvttaw0Go0GRERk8axFByAiImlgIRAREQAWAhER3cVCICIiACwEIiK6q57oAHVVXl6OU6dOwdHRETKZTHQcIiKToFKpcPXqVbi7u6NBgwbVjplsIZw6dQqjR48WHYOIyCRt2bIFXl5e1R4z2UJwdHQE8PcP1axZM8FpiIzru+++w6ZNm1BQUAB7e3vMnj0bHh4eomORCbhy5QpGjx6tew2tymQLQTtM1KxZM7i4uAhOQ2Q8//vf/xAfH4/ly5fDw8MDV69eBQA4OzsLTkam5EFD7byoTGRiVq5cialTp6J79+6wtraGs7Mzy+AR5efnY8CAAaJjSBYLgciEqFQqnDp1CiUlJQgMDMSAAQMQGxuL8vJy0dFMQlpaGvr37y86hmSxEIhMSGFhISorK5GcnIwtW7YgISEBp0+fxpo1a0RHMwmpqanw9fUVHUOyWAhEJkQ7TTA8PBxOTk5wcHDA+PHjkZqaKjiZ9FVWViIjIwM+Pj6io0gWC4HIhDRp0gTNmjWDlZWV7rGqH9PDZWZmolOnTrC1tRUdRbJYCEQmJiwsDJs2bUJRURGuX7+ODRs24LnnnhMdS/I4XFQ7FgKRCbh+U4md35/H9ZtKTJ06FV27dkVQUBBCQkLQuXNnTJkyRXREyWMh1M5k70MgsiQpGZcQn3gaABDm9yyio6MRHR0tNpSJuH5TiR3f/ohypRKurq6i40gaC4HIBAR4t6r2f3p0KRmXsGHrXri27y46iuRxyIjIBDSxtUGY37NoYmsjOorJCfBuBXtcxoiwENFRJI+FQERmrYmtDQa/4Af/5/qJjiJ5HDIiIrP3xhtviI5gEniGQEREAFgIRER0l94KYdGiRfD390fHjh1x7tw53eMXLlzAiBEjEBQUhBEjRuDixYuPdIyIiIxLb4UQEBCALVu2oEWLFtUenz9/PkaNGoX9+/dj1KhRmDdv3iMdIyIi49JbIXh5eUGhUFR7rKioCKdPn0ZoaCgAIDQ0FKdPn0ZxcXGNx4iIyPgMOssoLy8Pzs7Oup15ZDIZnJyckJeXB41G89BjDg4OhoxFREQPwIvKREQEwMBnCAqFAvn5+VCpVJDJZFCpVCgoKIBCoYBGo3noMSIiMj6DniE0bdoUbm5uSExMBAAkJibCzc0NDg4ONR4jIiLjs9JoNBp9fKGPP/4YBw4cQGFhIezt7WFnZ4ekpCRkZ2cjKioKpaWlaNy4MRYtWoR27doBQI3HanP58mUEBAQgJSUFLi4u+vgRiIjMXk2vnXorBGNjIZiWw4cPY/PmzcjKyoJKpULbtm0xfPhwhIWFwdqal7KIjKWm106uZUQGt3jxYvzwww+YOnUqFi9ejIYNG+LcuXNYt24d9u/fj1WrVkEul4uOSWTx+NaMDCohIQG//fYbtm3bhoEDB6Jx48aQyWRwc3PD0qVL0b59e8TFxYmOSUTgGQIZ2OrVq7Fx40bI5XIsWbIECQkJcHJygre3Nzp37oy3334boaGhmD59uuioRBaPhUAG8+eff8LJyQkKhQKpqalIS0vDnj17AADjxo1Dp06d0KBBAzRp0gTFxcWcYUYkGAuBDKaoqAjOzs4AgHPnzqF///5o2rQpAKB///4AALVajdLSUtjZ2YmKSUR38RoCGYy9vT0KCgoAAB06dMDhw4dRVFSEoqIipKen4/bt21i2bBl8fX0508hERUREoF+/fujRoweCgoKwY8cO0ZHoCfAMgQymbdu2uHLlCgoKCuDr64sff/wRQ4YMQfPmzeHt7Y2tW7ciJCQEb7/9tuioVEeTJk3CJ598ArlcjuzsbIwdOxZubm5wd3cXHY3qgIVAenf9phIpGZcQ4N0KEydOxIcffojVq1cjMjISkZGRuj93584d1Ksnzafg888/j7i4OLRp00Z0FEl79tlndR9bWVnBysoKly5dYiGYKJ6nk96lZFxCfOJppGRcwogRI+Di4oKxY8fihx9+gFKpxO3bt3Ho0CEMHToUOTk5ouM+0IABA5Camio6hkmIjo5Gt27d8MILL8DR0RG+vr6iI1EdSfPtGZm0AO9W1f4/b948HDx4EGvXrsVvv/0GAPD09ERsbOx9GypJha+vLzZs2IDXXntNdBTJi46Oxty5c/HTTz/hxx9/5E2GJoyFQHrXxNYGYX7PVnssMDAQgYGBghI9vt69eyMiIgJlZWV4+umnRceRPJlMBi8vL+zZswdbt27F2LFjRUeiOuCQEdEDyOVyeHp64ujRo6KjmBSVSoVLly6JjkF1xEIgeghfX19eR3iA6zeV2Pn9eVy4lIukpCTcunULKpUK6enpSEpKQt++fUVHpDrikBHRQzz33HNcZ+kBtJMGbvq2wIGvt2L+/PlQq9Vo0aIF5syZg4CAANERqY5YCET3qDpt1tbWFmfPnkXHjh1Fx5KMqpMGxg7eLDgN6RMLgege2nfAwD/TT1kI/3jQpAEyD7yGQHSPAO9WGB/aGQHereDr64u0tDTRkYiMgmcIRPeo+g64R48e6Nevn+BERMbBMwSiGtSrVw+TJ08WHYPIKFgIREQEgIVARER3sRCIiAgAC4GIiO7iLCMiE7d3717Ex8fjwoULaNiwITp16oTJkyfDy8tLdDQyMSwEIhMWHx+PuLg4xMTEoF+/fqhfvz7S09ORkpLCQqDHxkIgMlE3btzAihUr8Mknn+D555/XPe7v7w9/f3+ByUzfsmXL0LRpU4wbN050FKNiIRCZqJ9++glKpdKk9pkwBcXFxUhISMDBgwdFRzE6XlQmMlHXrl2Dvb29ZPelNlU7d+6Er68vGjRoIDqK0bEQiEyUnZ0dSkpKcOfOHdFRzEpaWhq8vb1FxxCChUBkojw9PSGXy3Ho0CHRUczKuXPn0LZtW9ExhGAhEJkY7Y5lais53nrrLcTGxuLQoUO4ffs2KisrkZqaisWLF4uOabJu3LiBhg0bio4hBAcfiUxM1f0aJkyYgGeeeQarV69GREQEGjZsiC5dunBBvifQuHFj3Lp1S3QMIVgIRCam6o5lADB48GAMHjxYZCSz0rFjR1y8eBEeHh6ioxgdh4yITIx2v4Ymtjaio5glX19fZGRkiI4hBAuBiAj/XJvxH/gCUlNTUV5eLjqS0bEQzIi/vz88PDzg6ekJHx8fREVFWexYKNHj0l6bOf77DQwdOhTbtm0THcnoeA3BzKxduxY+Pj7Iz8/H66+/jjVr1iAiIkJ0LLJQFRUViI6OxtGjR3Ht2jW0atUK77zzDnx9fUVHu0/VazNN/N4RnEYMFoKZcnZ2Rv/+/XH+/HnRUciC3blzBwqFAps2bULz5s2RmpqKmTNnYu/evXBxcREdr5qqe2lbKg4Zmam8vDykpaXBzc1NdBSyYE8//TRmzJgBFxcXWFtbw8/PDy4uLvj1119FR6MHMMoZgr+/P+RyOWxs/p4VERERgf79++PkyZOYN28elEolWrRogSVLlqBp06bGiGS2pk2bBplMhkaNGsHX19ek56NPmjQJx48ff+Cxnj174vPPPzdyInpShYWFuHjxItq3by86Cj2A0YaMVqxYgQ4dOug+V6vViIyMxMKFC+Hl5YXVq1dj6dKlWLhwobEimaVVq1bBx8dHdAy9kPIL/s6dOxEfH49Lly7B1tYWAwcOxLvvvovGjRuLjiZZlZWViIiIwLBhw+Dq6io6Dj2AsCGjU6dOwcbGRreJx8iRI5GcnCwqDtEjW79+PZYuXYrIyEhkZmZi+/btyM3Nxfjx41FRUSE6niSp1WrMnj0b9evXx9y5c0XHoYcw2hlCREQENBoNevbsiXfeeQd5eXlo3ry57riDgwPUajWuXbsGOzs7Y8UyC9dvKpGScQlqtUZ0FL2aOHFijUNGX3zxhZETATdv3sTKlSuxYMECDBgwAADg4uKCzz77DAEBAdizZw9eeuklo+eSMo1Ggw8++ACFhYX473//i/r164uORA9hlELYsmULFAoFKioqsGDBAsTGxnJTDz3Szp++rTSvZZBFvODX5sSJE1AqldV2KAOAhg0bwtfXF0eOHGEh3GP+/PnIzs5GfHy8Re4xYEqMMmSkUCgAAHK5HKNGjcKJEyegUCiQm5ur+zPFxcWwtrbm2UEdBHi3wvjQzjhw8JBZXD/Q3jF6/aZSdJT7lJSUPHRTGkdHR5SUlAhIJT3av8Pfzl/A9u3bcebMGfTr1w+enp7w9PTEnj17REekBzD4GUJZWRlUKhUaNWoEjUaDffv2wc3NDe7u7igvL0dmZia8vLywbds2BAcHGzqOWTK3+dNVV/OU2s9lb2+v25Tm3lK4evUq7O3tBSWTFu3f4fjQzjh79qzoOPSIDF4IRUVFmDFjBlQqFdRqNVxdXTF//nxYW1tj8eLFmD9/frVpp0T3ruYpJdpNaQ4cOICQkBDd47du3UJaWhreeccy73C9l5T/DunhDF4ILVu2REJCwgOP9ejRA3v37jV0BDIxUjzj0V64D/BuhWnTpuHjjz+Gra0t+vbti/z8fMTExKBZs2YYMmSI6KiSIMW/Q6odl64gegRVh7HeeOMN2NnZYfHixdXuQ1i6dCnkcrngpER1x0IgegT3DoG8/PLLePnll0VGItI7rmVE9AhMYVOasWPHomvXrnj11VdFRyED087iGjd+AtauXau3r8tCIDITX375JWJiYkTHICPQDmEOHvu+Xtcr45AREZGJMdQsLhYCEZGJMdQsLg4ZERERABYCERHdZdGFsHnzZoSFhcHd3R1RUVGi4xARCWXRheDk5ISpU6di+PDhoqMQ1YmUFwIk02PRF5W1SxhnZWUhPz9fcBqixyflhQDJ9Fh0IRCZOi4iR/rEQiAyYVWnH44fPx4nT56Eh4eH4FRkqlgIRGYiPj5edAQycRZ9UZmILEvHjh3RvXt3LF++XHQUSbLIMwTt2va+ns1h+1Q9qNVqqFQqKJVKyGSyB26PSETmYffu3WjdurXoGJJkka982pkZh/ZuQUrSVt3je/bswfTp0zFjxgyB6YhMi7+/PwoLC3Vvpjw9PRETE6PbS51Mh0UWwj8zM4LR5F/RYsMQmYG1a9fCx8cHSqUS0dHR+Oijj7B69WrRsegxWeQ1BFNY2570a+XKlVi5cqXoGGbPxsYGwcHByM7OFh2F6sAiC4GIDOP27dvYt28funXrJjoK1YFFDhmR/ly+fBkxMTE4efIk5HI5goKCMGfOHF6Yv0fVcXat5ORkODs7C0ylP9OmTYNMJsPt27dhb2+PdevWiY5EdcB/tfREYmJi0LRpUxw+fBilpaWYMGECvvrqK4wdO1Z0NMnRjrObo1WrVsHHxwcqlQopKSkIDw9HUlISHB0dRUejx8BCoCdy+fJljBkzBjY2NnB0dES/fv3w+++/i44FAJg0aRKOHz8OAFAq/178bePGjQCAnj174vPPPxeWzVzJZDI8//zzmDdvHo4fP47g4GDRkQD8M9WcasZCoCfy2muvISkpCb169UJpaSnS09Px9ttvi44FANVe8LUXlDmlWH+0L7JqtUb3mEajQUpKCkpLS+Hq6iowXXVVFwGkh2Mh0BPx9vbG119/jZ49e0KlUmHYsGEYOHCgsDzaF6kA71aSm0WmHWcHgF69epn8tEzti+xt5R1MnjxZ97O1aNECn376KZ59Vjqrr2qnms8/IEdYWBjCw8Mxc+ZMsaEkiIVAdaZWqzFx4kS88sor2LZtG27duoU5c+ZgyZIlmD17tpBMUl4OWjvObi509/PEHJJc+d5LO9U8LCtLdBRJYyFQnV27dg25ubkYM2YM5HI55HI5hg8fjs8++0xYITxsOWgOFemfoTZ6J3F4HwLVmYODA1xcXLB161bcuXMHpaWl2LVrFzp27Cgsk9RuOtTuaFZ1nJ2oLjIzMzFy5Ej07NkTvXr1wsiRI/HLL7/o9XuwEOixVd228T//+Q/S09PRt29fBAYGol69enj//fdFR5SMquPsRHV18+ZNTJ48GWPGjMGPP/6ItLQ0TJ8+HXK5XK/fh0NG9Niqj9O7YdOmTYITSZcpjbOTdF24cAEAEBoaCuDv6b39+vXT+/dhIdBj47aNj47j7KQPbdu2hUwmw3vvvYeQkBB0794dTZo00fv34ZARPTapjdMTmTtbW1t89dVXsLKywty5c9G3b19MnjwZhYWFev0+LAQiIhPg6uqKTz/9FGlpadi7dy8KCgrwySef6PV7sBCIiEyMq6srwsLCcP78eb1+XV5DICKSsOs3ldi6939QX/sdYcMGo1mzZsjLy0NiYqLelxlnIRARSVhKxiUkpF+GbckxbP1qE27cuIFGjRrBz89P7zeAshCIiCTsn1l9Iww+kYPXEIjM2J49e+Dp6QlPT09MnDhRdByqA2PO6hN+hnDhwgVERUXh2rVrsLOzw6JFi9CmTRvRsYjMwuDBgzF48GDRMchECD9DmD9/PkaNGoX9+/dj1KhRmDdvnuhIRCQRd+5wyQ9jEloIRUVFOH36tO527NDQUJw+fRrFxcUiY0nKSy+9pPepZWT+qq43ZWr8/f0RFxeHQYMGoXv37iwFIxJaCHl5eXB2dtZtrCGTyeDk5IS8vDyRsSRlwoQJWLFihegYRufv748jR46IjmGytOtNmeq2kUlJSYiLi0NmZibq1RM+sm0xhA8ZUc0CAgJw7NgxXL16VXQUi+fv74++ffuirKxM99iOHTsQHh4uMNWDBXi3wvjQzia73lR4eDgUCgUaNGggOopFEVoICoUC+fn5UKlUAACVSoWCggIoFAqRsSTFxsYGXbp0weHDh0VHIfy9S9yXX34pOkatTH29Kb4GiCG0EJo2bQo3NzckJiYCABITE+Hm5gYHBweRsSTH1dUVv/32m+gYBOD111/H+vXrUVpaKjqKWbOyshIdwSIJHzKKjo7G5s2bERQUhM2bNyMmJkZ0JMlp2LAhX4Akwt3dHb169cK6detERyHSO+FXa1xdXbFjxw7RMSTt1q1baNy4segYdNdbb72FV199Fa+99proKGbl+k0lUjIucbtRgYSfIVDtsrOz0alTJ9Ex6K4OHTrgueeeQ1xcnOgoZkU7M+qt+XHw8fERHccisRAkqOoccqVSiV9//ZX/QCTmrbfewtdff438/HzRUcyGqc+MMgfCh4zoflX3LH6q/Hf06tULzs7OglNRVa1bt0ZISAg2bdqEDh06iI5jFrjdqHgsBAmqumfxxPFzsGDBAsGJjEeK48gPyzRt2jTs3r1bUCoi/eOQkQRVnUO+Y8cOi3oHKsVx5IdlUigUyMrKwqZNmwSmI2N70F30O3fuxKuvviookf7wDIEkperZkVRIMRORIbAQSFKkOI4sxUxEhsAhIyIiAsAzBCKixzZt2jTdKs0AUFlZic6dOwtMpB8sBCKix7Rq1apqEwx27txpFisucMiISOLi4uIwcOBAeHp6IiQkBAcPHhQdicwUzxCIJK5ly5bYsmULHB0dkZycjMjISBw4cABOTk6io1kUKd4jo288QyCSuBdeeAHOzs6wtrZGSEgIWrdujV9++UV0LJ2XXnoJFy5cwF9//YVhw4aJjmMw2vtRbivNd0tPniEQSVxCQgLi4+ORk5MDACgrK0NJSYngVH+rrKxEbm4u2rRpg+TkZLO4sPowuvtRYg7dt/FQWFgYwsLCRMTSKxYCkYTl5OTgww8/xIYNG+Dp6QmZTIYhQ4aIjqVz/vx5uLq6wsrKCqdOnTLrQrCE+1FYCEQSpB2vbtu0ElZWVrpdBL/55hucP39ecLq/cyxcuBCVlZVQq9Xw8vJCWVkZbGxssHz5cuzatQstW7YUHZMeE68hEEmQdrz6QlF9TJgwASNHjoSPjw/OnTuHHj16iI6H4cOHIzMzE126dMHXX3+NPXv24Nlnn8WJEyeQmZnJMjBRPEMgkqCq6yc18ZuFWbNmCU70j2vXrmHgwIHQaDQoKytDeHg4KioqAADe3t6YPn06xo0bJzYk1QkLgUiCpDxebWdnh8zMTCQlJeHYsWOIjY3FtGnTMHr0aMmsUEt1wyEjqpG/vz/WrVuHQYMGoWfPnpg5cyaUSqXoWCRA1Z38AFS7iHzmzBl06dJFZDzSAxYC1erbb7/FF198gZSUFJw9exY7d+4UHemxvPjiizh27JjoGCZPe10jJeMSAODXX39F586dUVJSAmtrazRp0kRwQnpSHDKiWoWHh+u28PTz88OZM2cEJ3o8SUlJoiOYhXv3hfjyyy91xw4dOiQkE+kXzxCoVo6OjrqPn3rqKZSVlQlMQ6JU3cmPzBMLgczeg7Y8JKL7sRDoPvdePCQiy8BrCHQf7cVDIrIsLAS6T9WLh2HffVft2IwZM0REIiIjYCHQfaR8UxQRGQ4LgcyWJWxoQqRPvKhMZssSNjQh0ieeIZDZqmlDEyK6H88QDCguLg79+/eHp6cngoKCcPToUdGRLApvpDIPvI/EeHiGYCB//PEHtmzZgv/7v/+Ds7MzLl++DLVaLToWEdFDsRAMRCaToaKiAtnZ2XBwcICLi4voSERENeKQkYG0bt0ac+bMwcqVK+Hj44NZs2YhPz9fdCwioodiIRjQoEGDsHXrVnz//fewsrLC0qVLRUciicjPz8eMGTPQp08f+Pv7V1s5lEgUFoKeadcB+uXUbzh69CgqKiogl8thY2MDa2v+uglQq9WYMmUKOnbsiLS0NGzcuBEbN25Eenq66Ghk4XgNQc+0c9+Duz+F73fHITs7G/Xr14enpydiY2NFxyMJyMrKQnFxMaZPnw4AaNmyJV555RXs27cP/fv3F5yOLBkLQc+qrgM0Lfx5wWlIinJyclBQUAAvLy/dYyqVqtrnRCIYtBCioqJw5MgR2NvbAwCCg4MxZcoUAEBhYSFmz56NnJwc2NjY4KOPPkK3bt0MGccouA4Q1UahUMDFxQUHDhwQHUXSuPSI8Rn8DOHNN9/EmDFj7nt82bJl8PLywvr165GZmYnIyEjs378fVlZWho5EJJSHhwcaNmyIuLg4jB07FvXr10d2djbKy8vh4eEhOp5kaIdf35ofBx8fvskyBmFDRsnJyUhJSQEAeHl5QS6XIysri/8gyKxp3/Uu+dcKrF65HAEBAaioqEDbtm0xc+ZM0fEk5d49nMnwDF4I8fHx2L59O1q2bIl3330Xrq6uKCkpgUajgYODg+7PKRQKXLlyhYVAZk37rnd8aGf861//Eh1H0jj8anxPVAjDhg1Dbm7uA48dOXIEs2bNgqOjI6ytrZGQkICJEyfi0KFDT/ItiUwa3/WSlD1RIezatavG487OzrqPhw4dioULF+LKlSto0aIFAKC4uFh3lpCXl4dmzZo9SRwiyeO7XpIyg94pVXWphvT0dFhbW+tKIjg4GNu2bQMAZGZmory8HO7u7oaMQ0Z08eJFdO3aFREREaKjENEjMug1hPfeew9FRUWwsrKCra0t1qxZg3r1/v6W7777LiIjI5GQkAAbGxssXryYd/KakdjYWHTt2lV0DCJ6DAYthA0bNjz0mKOjY43HyXQlJSWhUaNG8PT0xJ9//ik6DhE9Ir4lJ726efMmVqxYgffff190FCJ6TFy6gvTqs88+w/DhwzlBwIR4enpW+7y8vByjRo3C3LlzBSUiUVgIpDdnzpzB0aNHa519RtLy008/6T6+desW+vXrh+DgYIGJSBQWAunNsWPHkJOTAz8/PwBAWVkZVCoVhg0bxpIwEQcOHICDgwMX2rNQLATSmxEjRuDFF1/Ufb5+/Xrk5OQgOjpaXCh6LLt27cLQoUO5ppiF4kVlemLaTYEqVNZwdHTU/ff0009DLpdXW6KEpCsnJwcZGRkYOnSo6CgkCM8Q6Ilp1+cBUO0u3BkzZoiKRHWwe/du9OzZEy1bthQdhQRhIdAT4/o85mH37t144403RMcggVgI9MS4Po/pO3HiBPLz8zm7yMLxGoIJyM7ORnh4OLy8vPDiiy/q9pEgehLaaz/XbyqRkJCAwMBA2Nraio5FAvEMQeIqKysxefJkDB8+HOvWrcPx48cxdepUfPPNN2jXrp3oeGTCql77iY2NFZyGpIBnCBL3888/o6ysDG+++Sbkcjn69u0LPz8/JCUliY5GJi7AuxXGh3bmtR/S4RmCxBUUFKBZs2bVVoJt3rx5taXFieqC137oXjxDkDgnJydcuXIFarVa91heXl61zYeIiPSBhSBxHh4eaNCgAb744gtUVlbi2LFj+O677xASEiI6GhGZGRaCRGlngNyu0GDt2rVIS0tDnz59EBMTg8WLF8PV1VV0RCIyM7yGIFH33v27efNmwYmIyNyxECSKd/8aRnh4OE6ePKnbytXJyQn79+8XnIpIGlgIEsUZIIYzb948vPzyy6JjEEkOryEQEREAFgJZoGXLlqF3794YOXIkjh07JjoOkWRwyIgsSkREBFxdXSGXy5GUlITJkydj9+7daNWK12qIeIZAFqVbt26wtbWFXC7HsGHD0KNHD6SmpoqORSQJLASyaFZWVtBoNKJjEEkCC4HMnvYmv8t5V5Geng6lUok7d+5gz549yMzMRP/+/UVHJJIEXkMgs6e9ye/mjRbYvfEz/PHHH5DJZGjXrh1WrVqFtm3bio5IJAksBDJ7VW/yGzv4G8FpiKSLhUBmjzf5ET0aXkMgIiIALAQiIrqLhUBERABYCEREdBcLgR5JXl4epk+fjj59+qB3796IjY0VHYmI9IyzjKhWKpUKkyZNQp8+ffDdd99BJpMhKytLdCwi0jMWAtXql19+QUFBAWbPnq3bWMbLy0twKiLSNw4ZUa3y8vLQvHlzXRkQkXliIVCtFAoF8vLycOfOHdFRiMiAWAj0UNpF4Vq36whHR0csW7YMZWVlUCqVOH78uOh4RKRnT1wIu3fvxqBBg9C5c2ds3ry52rHbt29j5syZCAwMRHBwML7//vtHOkbSoF0U7v+dyMHatWvx559/ws/PDwMGDMC3334rOh4R6dkTDwq7ublh+fLliIuLu+/YunXrYGtri4MHD+LixYsYPXo0Dhw4gIYNG9Z4jKSh6qJwTWxtsHr1asGJiMiQnvgMoUOHDmjfvj2sre//Ut9++y1GjBgBAGjTpg3c3d2RlpZW6zGSBu2icE1sbURHISIjMOg1hNzcXLRo0UL3uUKhwJUrV2o9RkRExlfrkNGwYcOQm5v7wGNHjhyBTCbTeygiIjK+Wgth165ddf7izZs3R05ODhwcHAD8PZ+9d+/etR4jIiLjM+iQUXBwMLZv3w4AuHjxIrKysnT719Z0jIiIjO+JCyExMREDBgxAcnIy/v3vf2PAgAH4/fffAQCvv/46SktLERgYiEmTJiE2Nha2tra1HiMiIuOz0mg0GtEh6uLy5csICAhASkoKXFxcRMchIjIJNb12muziNCqVCgA4M4mI6DFoXzO1r6FVmWwhXL16FQAwevRowUmIiEzP1atX0bp162qPmeyQUXl5OU6dOgVHR0dOfSUiekQqlQpXr16Fu7s7GjRoUO2YyRYCERHpF1c7JSIiACwEIiK6i4VAREQAWAhERHQXC4GIiACwEIiI6C4WAhERAbDAQuAe0A8XFRWFAQMGYMiQIRgyZAjWrFmjO1ZYWIgJEyYgKCgIgwcPxs8//ywwqXFcuHABI0aMQFBQEEaMGIGLFy+KjmRU/v7+CA4O1j0f0tPTAQAnT57E4MGDERQUhAkTJqCoqEhwUv1btGgR/P390bFjR5w7d073eE3PCbN4vmgszNmzZzXnz5/XREZGajZt2lTt2MqVKzUffPCBRqPRaC5cuKDx8fHR3Lx5s9Zj5uK9996773eiFRUVpVm1apVGo9FoMjIyNIGBgRq1Wm3MeEYXHh6uSUhI0Gg0Gk1CQoImPDxccCLj8vPz05w9e7baYyqVSjNw4EBNRkaGRqPRaFatWqWJiooSEc+gMjIyNLm5uff9Dmp6TpjD88XizhC4B3TdJCcnY+TIkQAALy8vyOVyZGVlCU5lOEVFRTh9+jRCQ0MBAKGhoTh9+jSKi4sFJxPr1KlTsLGxgZeXFwBg5MiRSE5OFpxK/7y8vKBQKKo9VtNzwlyeLxZXCDXhHtBAfHw8Bg0ahKlTpyI7OxsAUFJSAo1Go9vdDjDfn18rLy8Pzs7OunWyZDIZnJyckJeXJziZcUVERGDQoEGIjo5GaWkp8vLy0Lx5c91xBwcHqNVqXLt2TVxII6npOWEuzxeTXe30YbgH9MPV9ruZNWsWHB0dYW1tjYSEBEycOBGHDh0yckqSii1btkChUKCiogILFixAbGwsAgMDRcciAzK7QuAe0A9X2+/G2dlZ9/HQoUOxcOFCXLlyRXdmVFxcXO3nb9asmeHCCqZQKJCfnw+VSgWZTAaVSoWCgoL7hhHMmfZnlcvlGDVqFKZMmYKxY8dWe1NRXFwMa2tr2NnZCUppPDU9JzQajVk8XzhkVIWl7wGdn5+v+zg9PR3W1ta6kggODsa2bdsAAJmZmSgvL4e7u7uQnMbQtGlTuLm5ITExEcDfW8W6ublVGzYzZ2VlZbhx4wYAQKPRYN++fXBzc4O7uzvKy8uRmZkJANi2bRuCg4NFRjWamp4T5vJ8sbjlrxMTE7F48WKUlpaifv36eOqpp7B+/Xq0b98eZWVliIqKwpkzZ2BtbY3IyEgMHDgQAGo8Zi7GjRuHoqIiWFlZwdbWFrNnz0b37t0B/L2ZRmRkJHJzc2FjY4OYmBj06NFDbGADy87ORlRUFEpLS9G4cWMsWrQI7dq1Ex3LKP766y/MmDEDKpUKarUarq6u+PDDD+Hk5IQTJ05g/vz5UCqVaNGiBZYsWYJnnnlGdGS9+vjjj3HgwAEUFhbC3t4ednZ2SEpKqvE5YQ7PF4srBCIiejAOGREREQAWAhER3cVCICIiACwEIiK6i4VAREQAWAhERHQXC4GIiACwEIiI6K7/D9xfTo9T7bA4AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(X_2d[:, 0], X_2d[:, 1], s=1)\n",
    "for i, t in enumerate(tk.word_index):\n",
    "    plt.text(X_2d[i, 0], X_2d[i, 1], t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3f9f2cc6-8687-4676-b1aa-437914379f12",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}