1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "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>structureId</th>\n",
       "      <th>chainId</th>\n",
       "      <th>sequence</th>\n",
       "      <th>residueCount</th>\n",
       "      <th>macromoleculeType</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>100D</td>\n",
       "      <td>A</td>\n",
       "      <td>CCGGCGCCGG</td>\n",
       "      <td>20</td>\n",
       "      <td>DNA/RNA Hybrid</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>100D</td>\n",
       "      <td>B</td>\n",
       "      <td>CCGGCGCCGG</td>\n",
       "      <td>20</td>\n",
       "      <td>DNA/RNA Hybrid</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101D</td>\n",
       "      <td>A</td>\n",
       "      <td>CGCGAATTCGCG</td>\n",
       "      <td>24</td>\n",
       "      <td>DNA</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>101D</td>\n",
       "      <td>B</td>\n",
       "      <td>CGCGAATTCGCG</td>\n",
       "      <td>24</td>\n",
       "      <td>DNA</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>101M</td>\n",
       "      <td>A</td>\n",
       "      <td>MVLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDR...</td>\n",
       "      <td>154</td>\n",
       "      <td>Protein</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  structureId chainId                                           sequence  \\\n",
       "0        100D       A                                         CCGGCGCCGG   \n",
       "1        100D       B                                         CCGGCGCCGG   \n",
       "2        101D       A                                       CGCGAATTCGCG   \n",
       "3        101D       B                                       CGCGAATTCGCG   \n",
       "4        101M       A  MVLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDR...   \n",
       "\n",
       "   residueCount macromoleculeType  \n",
       "0            20    DNA/RNA Hybrid  \n",
       "1            20    DNA/RNA Hybrid  \n",
       "2            24               DNA  \n",
       "3            24               DNA  \n",
       "4           154           Protein  "
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "data = pd.read_csv('pdb_data_seq.csv')\n",
    "data.head()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(432474, 5)\n",
      "structureId          0\n",
      "chainId              0\n",
      "sequence             0\n",
      "residueCount         0\n",
      "macromoleculeType    0\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "data = data.dropna()\n",
    "print(data.shape)\n",
    "print(data.isnull().sum())\n",
    "data = data.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(432474, 2)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "new_data = []\n",
    "for i in range(len(data)):\n",
    "    if data[i][4] == '':\n",
    "        continue\n",
    "    elif data[i][4] == 'DNA' or data[i][4]== 'Protein'or data[i][4] == 'RNA':\n",
    "        new_data.append([data[i][2],data[i][4]])\n",
    "    else:\n",
    "        new_data.append([data[i][2],'Hybrid'])\n",
    "        \n",
    "new_data = np.array(new_data)\n",
    "new_data.shape\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>sequence</th>\n",
       "      <th>type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>CCGGCGCCGG</td>\n",
       "      <td>Hybrid</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>CCGGCGCCGG</td>\n",
       "      <td>Hybrid</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>CGCGAATTCGCG</td>\n",
       "      <td>DNA</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>CGCGAATTCGCG</td>\n",
       "      <td>DNA</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>MVLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDR...</td>\n",
       "      <td>Protein</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            sequence     type\n",
       "0                                         CCGGCGCCGG   Hybrid\n",
       "1                                         CCGGCGCCGG   Hybrid\n",
       "2                                       CGCGAATTCGCG      DNA\n",
       "3                                       CGCGAATTCGCG      DNA\n",
       "4  MVLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDR...  Protein"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.DataFrame(new_data, columns=['sequence', 'type'])\n",
    "data.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelBinarizer\n",
    "\n",
    "# Transform labels to one-hot\n",
    "lb = LabelBinarizer()\n",
    "Y = lb.fit_transform(data.type)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/smart_open/smart_open_lib.py:398: UserWarning: This function is deprecated, use smart_open.open instead. See the migration notes for details: https://github.com/RaRe-Technologies/smart_open/blob/master/README.rst#migrating-to-the-new-open-function\n",
      "  'See the migration notes for details: %s' % _MIGRATION_NOTES_URL\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Word2Vec(vocab=9414, size=300, alpha=0.025)\n"
     ]
    }
   ],
   "source": [
    "from gensim.models import Word2Vec\n",
    "new_model = Word2Vec.load('model.bin')\n",
    "print(new_model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def featureVecMethod(words, model, num_features):\n",
    "    # Pre-initialising empty numpy array for speed\n",
    "    featureVec = np.zeros(num_features,dtype=\"float32\")\n",
    "    nwords = 0\n",
    "    \n",
    "    #Converting Index2Word which is a list to a set for better speed in the execution.\n",
    "    index2word_set = set(model.wv.index2word)\n",
    "    \n",
    "    for word in  words:\n",
    "        if word in index2word_set:\n",
    "            nwords = nwords + 1\n",
    "            featureVec = np.add(featureVec,model[word])\n",
    "    \n",
    "    # Dividing the result by number of words to get average\n",
    "    featureVec = np.divide(featureVec, nwords)\n",
    "    return featureVec\n",
    "\n",
    "def getAvgFeatureVecs(sequence, model, num_features):\n",
    "    counter = 0\n",
    "    SeqFeatureVecs = np.zeros((len(sequence),num_features),dtype=\"float32\")\n",
    "    for seq in sequence:\n",
    "\n",
    "        if counter%1000 == 0:\n",
    "            print(\"Sequence %d of %d\"%(counter,len(sequence)))\n",
    "            \n",
    "        SeqFeatureVecs[counter] = featureVecMethod(seq, model, num_features)\n",
    "        counter = counter+1\n",
    "        \n",
    "    return SeqFeatureVecs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_data = []\n",
    "raw_label = []\n",
    "for a in data.values:\n",
    "    b = []\n",
    "    tmp = a[0]\n",
    "    tmp_lab = a[1]\n",
    "    for i in range(0, len(tmp), 3):\n",
    "        b.append(tmp[i:i+3])\n",
    "    raw_data.append(b)\n",
    "    raw_label.append(tmp_lab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(raw_data, Y, test_size=.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "345979"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequence 0 of 345979\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:12: DeprecationWarning: Call to deprecated `__getitem__` (Method will be removed in 4.0.0, use self.wv.__getitem__() instead).\n",
      "  if sys.path[0] == '':\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequence 1000 of 345979\n",
      "Sequence 2000 of 345979\n",
      "Sequence 3000 of 345979\n",
      "Sequence 4000 of 345979\n",
      "Sequence 5000 of 345979\n",
      "Sequence 6000 of 345979\n",
      "Sequence 7000 of 345979\n",
      "Sequence 8000 of 345979\n",
      "Sequence 9000 of 345979\n",
      "Sequence 10000 of 345979\n",
      "Sequence 11000 of 345979\n",
      "Sequence 12000 of 345979\n",
      "Sequence 13000 of 345979\n",
      "Sequence 14000 of 345979\n",
      "Sequence 15000 of 345979\n",
      "Sequence 16000 of 345979\n",
      "Sequence 17000 of 345979\n",
      "Sequence 18000 of 345979\n",
      "Sequence 19000 of 345979\n",
      "Sequence 20000 of 345979\n",
      "Sequence 21000 of 345979\n",
      "Sequence 22000 of 345979\n",
      "Sequence 23000 of 345979\n",
      "Sequence 24000 of 345979\n",
      "Sequence 25000 of 345979\n",
      "Sequence 26000 of 345979\n",
      "Sequence 27000 of 345979\n",
      "Sequence 28000 of 345979\n",
      "Sequence 29000 of 345979\n",
      "Sequence 30000 of 345979\n",
      "Sequence 31000 of 345979\n",
      "Sequence 32000 of 345979\n",
      "Sequence 33000 of 345979\n",
      "Sequence 34000 of 345979\n",
      "Sequence 35000 of 345979\n",
      "Sequence 36000 of 345979\n",
      "Sequence 37000 of 345979\n",
      "Sequence 38000 of 345979\n",
      "Sequence 39000 of 345979\n",
      "Sequence 40000 of 345979\n",
      "Sequence 41000 of 345979\n",
      "Sequence 42000 of 345979\n",
      "Sequence 43000 of 345979\n",
      "Sequence 44000 of 345979\n",
      "Sequence 45000 of 345979\n",
      "Sequence 46000 of 345979\n",
      "Sequence 47000 of 345979\n",
      "Sequence 48000 of 345979\n",
      "Sequence 49000 of 345979\n",
      "Sequence 50000 of 345979\n",
      "Sequence 51000 of 345979\n",
      "Sequence 52000 of 345979\n",
      "Sequence 53000 of 345979\n",
      "Sequence 54000 of 345979\n",
      "Sequence 55000 of 345979\n",
      "Sequence 56000 of 345979\n",
      "Sequence 57000 of 345979\n",
      "Sequence 58000 of 345979\n",
      "Sequence 59000 of 345979\n",
      "Sequence 60000 of 345979\n",
      "Sequence 61000 of 345979\n",
      "Sequence 62000 of 345979\n",
      "Sequence 63000 of 345979\n",
      "Sequence 64000 of 345979\n",
      "Sequence 65000 of 345979\n",
      "Sequence 66000 of 345979\n",
      "Sequence 67000 of 345979\n",
      "Sequence 68000 of 345979\n",
      "Sequence 69000 of 345979\n",
      "Sequence 70000 of 345979\n",
      "Sequence 71000 of 345979\n",
      "Sequence 72000 of 345979\n",
      "Sequence 73000 of 345979\n",
      "Sequence 74000 of 345979\n",
      "Sequence 75000 of 345979\n",
      "Sequence 76000 of 345979\n",
      "Sequence 77000 of 345979\n",
      "Sequence 78000 of 345979\n",
      "Sequence 79000 of 345979\n",
      "Sequence 80000 of 345979\n",
      "Sequence 81000 of 345979\n",
      "Sequence 82000 of 345979\n",
      "Sequence 83000 of 345979\n",
      "Sequence 84000 of 345979\n",
      "Sequence 85000 of 345979\n",
      "Sequence 86000 of 345979\n",
      "Sequence 87000 of 345979\n",
      "Sequence 88000 of 345979\n",
      "Sequence 89000 of 345979\n",
      "Sequence 90000 of 345979\n",
      "Sequence 91000 of 345979\n",
      "Sequence 92000 of 345979\n",
      "Sequence 93000 of 345979\n",
      "Sequence 94000 of 345979\n",
      "Sequence 95000 of 345979\n",
      "Sequence 96000 of 345979\n",
      "Sequence 97000 of 345979\n",
      "Sequence 98000 of 345979\n",
      "Sequence 99000 of 345979\n",
      "Sequence 100000 of 345979\n",
      "Sequence 101000 of 345979\n",
      "Sequence 102000 of 345979\n",
      "Sequence 103000 of 345979\n",
      "Sequence 104000 of 345979\n",
      "Sequence 105000 of 345979\n",
      "Sequence 106000 of 345979\n",
      "Sequence 107000 of 345979\n",
      "Sequence 108000 of 345979\n",
      "Sequence 109000 of 345979\n",
      "Sequence 110000 of 345979\n",
      "Sequence 111000 of 345979\n",
      "Sequence 112000 of 345979\n",
      "Sequence 113000 of 345979\n",
      "Sequence 114000 of 345979\n",
      "Sequence 115000 of 345979\n",
      "Sequence 116000 of 345979\n",
      "Sequence 117000 of 345979\n",
      "Sequence 118000 of 345979\n",
      "Sequence 119000 of 345979\n",
      "Sequence 120000 of 345979\n",
      "Sequence 121000 of 345979\n",
      "Sequence 122000 of 345979\n",
      "Sequence 123000 of 345979\n",
      "Sequence 124000 of 345979\n",
      "Sequence 125000 of 345979\n",
      "Sequence 126000 of 345979\n",
      "Sequence 127000 of 345979\n",
      "Sequence 128000 of 345979\n",
      "Sequence 129000 of 345979\n",
      "Sequence 130000 of 345979\n",
      "Sequence 131000 of 345979\n",
      "Sequence 132000 of 345979\n",
      "Sequence 133000 of 345979\n",
      "Sequence 134000 of 345979\n",
      "Sequence 135000 of 345979\n",
      "Sequence 136000 of 345979\n",
      "Sequence 137000 of 345979\n",
      "Sequence 138000 of 345979\n",
      "Sequence 139000 of 345979\n",
      "Sequence 140000 of 345979\n",
      "Sequence 141000 of 345979\n",
      "Sequence 142000 of 345979\n",
      "Sequence 143000 of 345979\n",
      "Sequence 144000 of 345979\n",
      "Sequence 145000 of 345979\n",
      "Sequence 146000 of 345979\n",
      "Sequence 147000 of 345979\n",
      "Sequence 148000 of 345979\n",
      "Sequence 149000 of 345979\n",
      "Sequence 150000 of 345979\n",
      "Sequence 151000 of 345979\n",
      "Sequence 152000 of 345979\n",
      "Sequence 153000 of 345979\n",
      "Sequence 154000 of 345979\n",
      "Sequence 155000 of 345979\n",
      "Sequence 156000 of 345979\n",
      "Sequence 157000 of 345979\n",
      "Sequence 158000 of 345979\n",
      "Sequence 159000 of 345979\n",
      "Sequence 160000 of 345979\n",
      "Sequence 161000 of 345979\n",
      "Sequence 162000 of 345979\n",
      "Sequence 163000 of 345979\n",
      "Sequence 164000 of 345979\n",
      "Sequence 165000 of 345979\n",
      "Sequence 166000 of 345979\n",
      "Sequence 167000 of 345979\n",
      "Sequence 168000 of 345979\n",
      "Sequence 169000 of 345979\n",
      "Sequence 170000 of 345979\n",
      "Sequence 171000 of 345979\n",
      "Sequence 172000 of 345979\n",
      "Sequence 173000 of 345979\n",
      "Sequence 174000 of 345979\n",
      "Sequence 175000 of 345979\n",
      "Sequence 176000 of 345979\n",
      "Sequence 177000 of 345979\n",
      "Sequence 178000 of 345979\n",
      "Sequence 179000 of 345979\n",
      "Sequence 180000 of 345979\n",
      "Sequence 181000 of 345979\n",
      "Sequence 182000 of 345979\n",
      "Sequence 183000 of 345979\n",
      "Sequence 184000 of 345979\n",
      "Sequence 185000 of 345979\n",
      "Sequence 186000 of 345979\n",
      "Sequence 187000 of 345979\n",
      "Sequence 188000 of 345979\n",
      "Sequence 189000 of 345979\n",
      "Sequence 190000 of 345979\n",
      "Sequence 191000 of 345979\n",
      "Sequence 192000 of 345979\n",
      "Sequence 193000 of 345979\n",
      "Sequence 194000 of 345979\n",
      "Sequence 195000 of 345979\n",
      "Sequence 196000 of 345979\n",
      "Sequence 197000 of 345979\n",
      "Sequence 198000 of 345979\n",
      "Sequence 199000 of 345979\n",
      "Sequence 200000 of 345979\n",
      "Sequence 201000 of 345979\n",
      "Sequence 202000 of 345979\n",
      "Sequence 203000 of 345979\n",
      "Sequence 204000 of 345979\n",
      "Sequence 205000 of 345979\n",
      "Sequence 206000 of 345979\n",
      "Sequence 207000 of 345979\n",
      "Sequence 208000 of 345979\n",
      "Sequence 209000 of 345979\n",
      "Sequence 210000 of 345979\n",
      "Sequence 211000 of 345979\n",
      "Sequence 212000 of 345979\n",
      "Sequence 213000 of 345979\n",
      "Sequence 214000 of 345979\n",
      "Sequence 215000 of 345979\n",
      "Sequence 216000 of 345979\n",
      "Sequence 217000 of 345979\n",
      "Sequence 218000 of 345979\n",
      "Sequence 219000 of 345979\n",
      "Sequence 220000 of 345979\n",
      "Sequence 221000 of 345979\n",
      "Sequence 222000 of 345979\n",
      "Sequence 223000 of 345979\n",
      "Sequence 224000 of 345979\n",
      "Sequence 225000 of 345979\n",
      "Sequence 226000 of 345979\n",
      "Sequence 227000 of 345979\n",
      "Sequence 228000 of 345979\n",
      "Sequence 229000 of 345979\n",
      "Sequence 230000 of 345979\n",
      "Sequence 231000 of 345979\n",
      "Sequence 232000 of 345979\n",
      "Sequence 233000 of 345979\n",
      "Sequence 234000 of 345979\n",
      "Sequence 235000 of 345979\n",
      "Sequence 236000 of 345979\n",
      "Sequence 237000 of 345979\n",
      "Sequence 238000 of 345979\n",
      "Sequence 239000 of 345979\n",
      "Sequence 240000 of 345979\n",
      "Sequence 241000 of 345979\n",
      "Sequence 242000 of 345979\n",
      "Sequence 243000 of 345979\n",
      "Sequence 244000 of 345979\n",
      "Sequence 245000 of 345979\n",
      "Sequence 246000 of 345979\n",
      "Sequence 247000 of 345979\n",
      "Sequence 248000 of 345979\n",
      "Sequence 249000 of 345979\n",
      "Sequence 250000 of 345979\n",
      "Sequence 251000 of 345979\n",
      "Sequence 252000 of 345979\n",
      "Sequence 253000 of 345979\n",
      "Sequence 254000 of 345979\n",
      "Sequence 255000 of 345979\n",
      "Sequence 256000 of 345979\n",
      "Sequence 257000 of 345979\n",
      "Sequence 258000 of 345979\n",
      "Sequence 259000 of 345979\n",
      "Sequence 260000 of 345979\n",
      "Sequence 261000 of 345979\n",
      "Sequence 262000 of 345979\n",
      "Sequence 263000 of 345979\n",
      "Sequence 264000 of 345979\n",
      "Sequence 265000 of 345979\n",
      "Sequence 266000 of 345979\n",
      "Sequence 267000 of 345979\n",
      "Sequence 268000 of 345979\n",
      "Sequence 269000 of 345979\n",
      "Sequence 270000 of 345979\n",
      "Sequence 271000 of 345979\n",
      "Sequence 272000 of 345979\n",
      "Sequence 273000 of 345979\n",
      "Sequence 274000 of 345979\n",
      "Sequence 275000 of 345979\n",
      "Sequence 276000 of 345979\n",
      "Sequence 277000 of 345979\n",
      "Sequence 278000 of 345979\n",
      "Sequence 279000 of 345979\n",
      "Sequence 280000 of 345979\n",
      "Sequence 281000 of 345979\n",
      "Sequence 282000 of 345979\n",
      "Sequence 283000 of 345979\n",
      "Sequence 284000 of 345979\n",
      "Sequence 285000 of 345979\n",
      "Sequence 286000 of 345979\n",
      "Sequence 287000 of 345979\n",
      "Sequence 288000 of 345979\n",
      "Sequence 289000 of 345979\n",
      "Sequence 290000 of 345979\n",
      "Sequence 291000 of 345979\n",
      "Sequence 292000 of 345979\n",
      "Sequence 293000 of 345979\n",
      "Sequence 294000 of 345979\n",
      "Sequence 295000 of 345979\n",
      "Sequence 296000 of 345979\n",
      "Sequence 297000 of 345979\n",
      "Sequence 298000 of 345979\n",
      "Sequence 299000 of 345979\n",
      "Sequence 300000 of 345979\n",
      "Sequence 301000 of 345979\n",
      "Sequence 302000 of 345979\n",
      "Sequence 303000 of 345979\n",
      "Sequence 304000 of 345979\n",
      "Sequence 305000 of 345979\n",
      "Sequence 306000 of 345979\n",
      "Sequence 307000 of 345979\n",
      "Sequence 308000 of 345979\n",
      "Sequence 309000 of 345979\n",
      "Sequence 310000 of 345979\n",
      "Sequence 311000 of 345979\n",
      "Sequence 312000 of 345979\n",
      "Sequence 313000 of 345979\n",
      "Sequence 314000 of 345979\n",
      "Sequence 315000 of 345979\n",
      "Sequence 316000 of 345979\n",
      "Sequence 317000 of 345979\n",
      "Sequence 318000 of 345979\n",
      "Sequence 319000 of 345979\n",
      "Sequence 320000 of 345979\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequence 321000 of 345979\n",
      "Sequence 322000 of 345979\n",
      "Sequence 323000 of 345979\n",
      "Sequence 324000 of 345979\n",
      "Sequence 325000 of 345979\n",
      "Sequence 326000 of 345979\n",
      "Sequence 327000 of 345979\n",
      "Sequence 328000 of 345979\n",
      "Sequence 329000 of 345979\n",
      "Sequence 330000 of 345979\n",
      "Sequence 331000 of 345979\n",
      "Sequence 332000 of 345979\n",
      "Sequence 333000 of 345979\n",
      "Sequence 334000 of 345979\n",
      "Sequence 335000 of 345979\n",
      "Sequence 336000 of 345979\n",
      "Sequence 337000 of 345979\n",
      "Sequence 338000 of 345979\n",
      "Sequence 339000 of 345979\n",
      "Sequence 340000 of 345979\n",
      "Sequence 341000 of 345979\n",
      "Sequence 342000 of 345979\n",
      "Sequence 343000 of 345979\n",
      "Sequence 344000 of 345979\n",
      "Sequence 345000 of 345979\n",
      "Sequence 0 of 86495\n",
      "Sequence 1000 of 86495\n",
      "Sequence 2000 of 86495\n",
      "Sequence 3000 of 86495\n",
      "Sequence 4000 of 86495\n",
      "Sequence 5000 of 86495\n",
      "Sequence 6000 of 86495\n",
      "Sequence 7000 of 86495\n",
      "Sequence 8000 of 86495\n",
      "Sequence 9000 of 86495\n",
      "Sequence 10000 of 86495\n",
      "Sequence 11000 of 86495\n",
      "Sequence 12000 of 86495\n",
      "Sequence 13000 of 86495\n",
      "Sequence 14000 of 86495\n",
      "Sequence 15000 of 86495\n",
      "Sequence 16000 of 86495\n",
      "Sequence 17000 of 86495\n",
      "Sequence 18000 of 86495\n",
      "Sequence 19000 of 86495\n",
      "Sequence 20000 of 86495\n",
      "Sequence 21000 of 86495\n",
      "Sequence 22000 of 86495\n",
      "Sequence 23000 of 86495\n",
      "Sequence 24000 of 86495\n",
      "Sequence 25000 of 86495\n",
      "Sequence 26000 of 86495\n",
      "Sequence 27000 of 86495\n",
      "Sequence 28000 of 86495\n",
      "Sequence 29000 of 86495\n",
      "Sequence 30000 of 86495\n",
      "Sequence 31000 of 86495\n",
      "Sequence 32000 of 86495\n",
      "Sequence 33000 of 86495\n",
      "Sequence 34000 of 86495\n",
      "Sequence 35000 of 86495\n",
      "Sequence 36000 of 86495\n",
      "Sequence 37000 of 86495\n",
      "Sequence 38000 of 86495\n",
      "Sequence 39000 of 86495\n",
      "Sequence 40000 of 86495\n",
      "Sequence 41000 of 86495\n",
      "Sequence 42000 of 86495\n",
      "Sequence 43000 of 86495\n",
      "Sequence 44000 of 86495\n",
      "Sequence 45000 of 86495\n",
      "Sequence 46000 of 86495\n",
      "Sequence 47000 of 86495\n",
      "Sequence 48000 of 86495\n",
      "Sequence 49000 of 86495\n",
      "Sequence 50000 of 86495\n",
      "Sequence 51000 of 86495\n",
      "Sequence 52000 of 86495\n",
      "Sequence 53000 of 86495\n",
      "Sequence 54000 of 86495\n",
      "Sequence 55000 of 86495\n",
      "Sequence 56000 of 86495\n",
      "Sequence 57000 of 86495\n",
      "Sequence 58000 of 86495\n",
      "Sequence 59000 of 86495\n",
      "Sequence 60000 of 86495\n",
      "Sequence 61000 of 86495\n",
      "Sequence 62000 of 86495\n",
      "Sequence 63000 of 86495\n",
      "Sequence 64000 of 86495\n",
      "Sequence 65000 of 86495\n",
      "Sequence 66000 of 86495\n",
      "Sequence 67000 of 86495\n",
      "Sequence 68000 of 86495\n",
      "Sequence 69000 of 86495\n",
      "Sequence 70000 of 86495\n",
      "Sequence 71000 of 86495\n",
      "Sequence 72000 of 86495\n",
      "Sequence 73000 of 86495\n",
      "Sequence 74000 of 86495\n",
      "Sequence 75000 of 86495\n",
      "Sequence 76000 of 86495\n",
      "Sequence 77000 of 86495\n",
      "Sequence 78000 of 86495\n",
      "Sequence 79000 of 86495\n",
      "Sequence 80000 of 86495\n",
      "Sequence 81000 of 86495\n",
      "Sequence 82000 of 86495\n",
      "Sequence 83000 of 86495\n",
      "Sequence 84000 of 86495\n",
      "Sequence 85000 of 86495\n",
      "Sequence 86000 of 86495\n"
     ]
    }
   ],
   "source": [
    "X_train = getAvgFeatureVecs(X_train, new_model, 300)\n",
    "X_test = getAvgFeatureVecs(X_test, new_model, 300)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_7 (Dense)              (None, 500)               150500    \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 500)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_5 (Batch (None, 500)               2000      \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 500)               250500    \n",
      "_________________________________________________________________\n",
      "dropout_6 (Dropout)          (None, 500)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_6 (Batch (None, 500)               2000      \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 300)               150300    \n",
      "_________________________________________________________________\n",
      "dropout_7 (Dropout)          (None, 300)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_7 (Batch (None, 300)               1200      \n",
      "_________________________________________________________________\n",
      "dense_10 (Dense)             (None, 4)                 1204      \n",
      "_________________________________________________________________\n",
      "activation_3 (Activation)    (None, 4)                 0         \n",
      "=================================================================\n",
      "Total params: 557,704\n",
      "Trainable params: 555,104\n",
      "Non-trainable params: 2,600\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers.recurrent import LSTM, GRU\n",
    "from keras.layers.core import Dense, Activation, Dropout\n",
    "from keras.layers.embeddings import Embedding\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.utils import np_utils\n",
    "\n",
    "from keras.layers import GlobalMaxPooling1D, Conv1D, MaxPooling1D, Flatten, Bidirectional, SpatialDropout1D\n",
    "\n",
    "model = Sequential()\n",
    "\n",
    "model.add(Dense(500, input_dim=300, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(BatchNormalization())\n",
    "\n",
    "model.add(Dense(500, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(BatchNormalization())\n",
    "\n",
    "model.add(Dense(300, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(BatchNormalization())\n",
    "\n",
    "model.add(Dense(4))\n",
    "model.add(Activation('softmax'))\n",
    "\n",
    "# compile the model\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 345979 samples, validate on 86495 samples\n",
      "Epoch 1/20\n",
      "345979/345979 [==============================] - 38s 110us/step - loss: 0.2691 - acc: 0.9122 - val_loss: 0.1639 - val_acc: 0.9434\n",
      "Epoch 2/20\n",
      "345979/345979 [==============================] - 30s 87us/step - loss: 0.1667 - acc: 0.9438 - val_loss: 0.1469 - val_acc: 0.9491\n",
      "Epoch 3/20\n",
      "345979/345979 [==============================] - 30s 87us/step - loss: 0.1470 - acc: 0.9499 - val_loss: 0.1257 - val_acc: 0.9547\n",
      "Epoch 4/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.1335 - acc: 0.9535 - val_loss: 0.1264 - val_acc: 0.9542\n",
      "Epoch 5/20\n",
      "345979/345979 [==============================] - 30s 87us/step - loss: 0.1244 - acc: 0.9563 - val_loss: 0.1208 - val_acc: 0.9562\n",
      "Epoch 6/20\n",
      "345979/345979 [==============================] - 30s 87us/step - loss: 0.1180 - acc: 0.9582 - val_loss: 0.1103 - val_acc: 0.9599\n",
      "Epoch 7/20\n",
      "345979/345979 [==============================] - 30s 87us/step - loss: 0.1129 - acc: 0.9606 - val_loss: 0.1115 - val_acc: 0.9598\n",
      "Epoch 8/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.1079 - acc: 0.9617 - val_loss: 0.1298 - val_acc: 0.9535\n",
      "Epoch 9/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.1051 - acc: 0.9628 - val_loss: 0.1086 - val_acc: 0.9601\n",
      "Epoch 10/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.1014 - acc: 0.9638 - val_loss: 0.1009 - val_acc: 0.9621\n",
      "Epoch 11/20\n",
      "345979/345979 [==============================] - 30s 86us/step - loss: 0.0983 - acc: 0.9649 - val_loss: 0.1081 - val_acc: 0.9603\n",
      "Epoch 12/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.0967 - acc: 0.9655 - val_loss: 0.1009 - val_acc: 0.9619\n",
      "Epoch 13/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.0937 - acc: 0.9662 - val_loss: 0.0978 - val_acc: 0.9640\n",
      "Epoch 14/20\n",
      "345979/345979 [==============================] - 31s 88us/step - loss: 0.0915 - acc: 0.9670 - val_loss: 0.1204 - val_acc: 0.9587\n",
      "Epoch 15/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.0896 - acc: 0.9678 - val_loss: 0.0897 - val_acc: 0.9668\n",
      "Epoch 16/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.0884 - acc: 0.9680 - val_loss: 0.1097 - val_acc: 0.9606\n",
      "Epoch 17/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.0869 - acc: 0.9683 - val_loss: 0.1083 - val_acc: 0.9596\n",
      "Epoch 18/20\n",
      "345979/345979 [==============================] - 31s 89us/step - loss: 0.0855 - acc: 0.9690 - val_loss: 0.0921 - val_acc: 0.9659\n",
      "Epoch 19/20\n",
      "345979/345979 [==============================] - 30s 88us/step - loss: 0.0843 - acc: 0.9696 - val_loss: 0.0920 - val_acc: 0.9663\n",
      "Epoch 20/20\n",
      "345979/345979 [==============================] - 31s 89us/step - loss: 0.0825 - acc: 0.9701 - val_loss: 0.0873 - val_acc: 0.9673\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=20, batch_size=128)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training - Accuracy = 97.3761413264967\n",
      "Test - Accuracy = 96.73044684663853\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib as plt\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.metrics import accuracy_score\n",
    "from sklearn.metrics import classification_report\n",
    "import itertools\n",
    "\n",
    "train_pred = model.predict(X_train)\n",
    "test_pred = model.predict(X_test)\n",
    "print(\"Training - Accuracy = \" + str(accuracy_score(np.argmax(y_train, axis=1), np.argmax(train_pred, axis=1))*100))\n",
    "print(\"Test - Accuracy = \" + str(accuracy_score(np.argmax(y_test, axis=1), np.argmax(test_pred, axis=1))*100))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "         DNA       0.79      0.76      0.77       783\n",
      "      Hybrid       0.96      0.86      0.91     16194\n",
      "     Protein       0.97      1.00      0.98     69069\n",
      "         RNA       0.71      0.76      0.73       449\n",
      "\n",
      "   micro avg       0.97      0.97      0.97     86495\n",
      "   macro avg       0.86      0.84      0.85     86495\n",
      "weighted avg       0.97      0.97      0.97     86495\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "# Compute confusion matrix\n",
    "cm = confusion_matrix(np.argmax(y_test, axis=1), np.argmax(test_pred, axis=1))\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "np.set_printoptions(precision=2)\n",
    "plt.figure(figsize=(10,10))\n",
    "plt.imshow(cm, interpolation='nearest', cmap=plt.cm.Blues)\n",
    "plt.title('Confusion matrix')\n",
    "plt.colorbar()\n",
    "tick_marks = np.arange(len(lb.classes_))\n",
    "plt.xticks(tick_marks, lb.classes_, rotation=90)\n",
    "plt.yticks(tick_marks, lb.classes_)\n",
    "#for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "#    plt.text(j, i, format(cm[i, j], '.2f'), horizontalalignment=\"center\", color=\"white\" if cm[i, j] > cm.max() / 2. else \"black\")\n",
    "plt.ylabel('True label')\n",
    "plt.xlabel('Predicted label')\n",
    "plt.show()\n",
    "\n",
    "print(classification_report(np.argmax(y_test, axis=1), np.argmax(test_pred, axis=1), target_names=lb.classes_))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "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": [
    "history_dict = history.history\n",
    "history_dict.keys()\n",
    "\n",
    "acc = history_dict['acc']\n",
    "val_acc = history_dict['val_acc']\n",
    "loss = history_dict['loss']\n",
    "val_loss = history_dict['val_loss']\n",
    "\n",
    "epochs = range(1, len(acc) + 1)\n",
    "\n",
    "plt.plot(epochs, loss, 'r', label='Training loss')\n",
    "# b is for \"solid blue line\"\n",
    "plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()\n",
    "\n",
    "plt.clf()   # clear figure\n",
    "\n",
    "plt.plot(epochs, acc, 'r', label='Training acc')\n",
    "plt.plot(epochs, val_acc, 'b', label='Validation acc')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedding_matrix = np.zeros((9415, 300))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(9415, 300)"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "for seq, i in tqdm(word_index.items()):\n",
    "    embedding_vector = embeddings_index.get(word)\n",
    "    if embedding_vector is not None:\n",
    "        embedding_matrix[i] = embedding_vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "9411"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "words = []\n",
    "for i in raw_data:\n",
    "    for j in i:\n",
    "        if j not in words:\n",
    "            words.append(j)\n",
    "len(words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}