LSTM vs FFNN only trained and tested on 9 mers .ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import sklearn \n",
    "import scipy\n",
    "from sklearn import linear_model as lm\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.decomposition import PCA\n",
    "from sklearn.cross_validation import KFold, train_test_split, cross_val_score, StratifiedKFold, LabelKFold, ShuffleSplit\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.svm import SVR\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from mhcflurry.amino_acid import common_amino_acids\n",
    "from mhcflurry import dataset\n",
    "from mhcflurry.dataset import Dataset\n",
    "import matplotlib.pyplot as plt \n",
    "% matplotlib inline\n",
    "import numpy as np\n",
    "import math \n",
    "from mhcflurry import peptide_encoding, amino_acid\n",
    "import statsmodels.api as sm\n",
    "from keras import models, layers, optimizers\n",
    "from keras.optimizers import Adam \n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Embedding, LSTM, Input, merge, Convolution1D, AveragePooling1D, Activation, Flatten\n",
    "from keras.preprocessing import sequence\n",
    "from keras.models import Model\n",
    "from keras.engine import topology\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "df = pd.read_table(\"bdata.2009.mhci.public.1.txt\")\n",
    "\n",
    "df['log_meas']=1-np.log(df['meas'])/math.log(50000)\n",
    "df['peptide_length'] = df['sequence'].str.len()\n",
    "\n",
    "\n",
    "max_len=df['sequence'].str.len().max()\n",
    "n_peptides = df['sequence'].count()\n",
    "\n",
    "def amino_acid_hotshot_encoding(s):\n",
    "    return common_amino_acids.hotshot_encoding([s],len(s)).flatten().astype(int)\n",
    "df['hotshot_encoded_peptides'] = df.sequence.apply(lambda seq: amino_acid_hotshot_encoding(seq))\n",
    "\n",
    "def amino_acid_index_encoding(s, maxlen):\n",
    "    a = 1+common_amino_acids.index_encoding([s],len(s)).flatten()\n",
    "    return np.concatenate([a, np.zeros(maxlen-len(a),dtype=int)])\n",
    "df['index_encoded_peptides'] = df.sequence.apply(lambda seq: amino_acid_index_encoding(seq,max_len))\n",
    "\n",
    "def measured_affinity_less_than(Y,k):\n",
    "    IC50 = 50000**(1-Y)\n",
    "    return (IC50 < k).astype(int) \n",
    "\n",
    "def affinity_label(Y):\n",
    "    return measured_affinity_less_than(Y,50) + measured_affinity_less_than(Y,500) + measured_affinity_less_than(Y,5000) + measured_affinity_less_than(Y,50000)\n",
    "\n",
    "df['affinity_label'] = affinity_label(df['log_meas'])\n",
    "df_h = df[df['mhc']=='HLA-A-0201'][['hotshot_encoded_peptides','index_encoded_peptides','log_meas','peptide_length']]\n",
    "X = np.array(list(df_h['index_encoded_peptides']))\n",
    "y = np.array(list(df_h['log_meas']))\n",
    "y[y<0]=0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "nine_idx = np.array([i for i in range(len(df_h)) if (np.count_nonzero(X[i])==9)])\n",
    "X_9 = X[nine_idx]\n",
    "y_9 = y[nine_idx]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(6961, 9565)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(X_9),len(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_9[1][9]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def delete_char_from_nine_onwards(X):\n",
    "    return np.delete(X, np.arange(9,max_len,1), None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "X_9_shortened = np.zeros((len(X_9),9))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "for i in range(len(X_9)):\n",
    "    X_9_shortened[i] = np.delete(X_9[i], np.arange(9,max_len,1), None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  1.,   1.,   1.,   9.,   1.,   1.,   1.,   1.,  18.])"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_9_shortened[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0717     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0469     \n",
      "All:  0.936376970085 0.938163501119 0 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0337     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0279     \n",
      "All:  0.952477242793 0.955766592042 1 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0267     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0246     \n",
      "All:  0.958178151018 0.956932393814 2 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0243     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0227     \n",
      "All:  0.960036040126 0.957599548811 3 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0231     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0225     \n",
      "All:  0.960943061416 0.957878962952 4 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0226     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0215     \n",
      "All:  0.961128764353 0.958391365501 5 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0221     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0212     \n",
      "All:  0.961322204913 0.9584945338 6 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0218     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0213     \n",
      "All:  0.961389264307 0.95857878791 7 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0217     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0210     \n",
      "All:  0.961710805504 0.958715485905 8 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0215     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0210     \n",
      "All:  0.961887051347 0.958768789526 9 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0214     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0211     \n",
      "All:  0.961449445814 0.958757612961 10 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0214     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0206     \n",
      "All:  0.96179591935 0.958654444662 11 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0212     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0206     \n",
      "All:  0.961415056381 0.958701730132 12 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0213     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0205     \n",
      "All:  0.961579265923 0.958703019736 13 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0212     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0203     \n",
      "All:  0.961480396304 0.958744716923 14 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0202     \n",
      "All:  0.96149759102 0.958727522207 15 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0210     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0206     \n",
      "All:  0.961553473849 0.958735259829 16 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0204     \n",
      "All:  0.961589582753 0.95876191164 17 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0210     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0202     \n",
      "All:  0.961245688425 0.958763631111 18 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0209     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0204     \n",
      "All:  0.961142520126 0.958783405035 19 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0209     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0206     \n",
      "All:  0.961400440873 0.958800599752 20 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0209     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0203     \n",
      "All:  0.961867277423 0.958806617902 21 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0208     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0203     \n",
      "All:  0.961570668565 0.958810916582 22 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0204     \n",
      "All:  0.961632569544 0.958821233411 23 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0204     \n",
      "All:  0.961758090974 0.958817794468 24 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0202     \n",
      "All:  0.961494152077 0.958816934732 25 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0204     \n",
      "All:  0.961715963919 0.958817794468 26 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0203     \n",
      "All:  0.961752932559 0.958828971034 27 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0205     \n",
      "All:  0.961961848363 0.958823812619 28 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0200     \n",
      "All:  0.962197415978 0.958845306014 29 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0203     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.96188017346 0.958841867071 30 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0202     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.962050401153 0.958852183901 31 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0202     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0200     \n",
      "All:  0.961908544742 0.958862500731 32 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0201     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0199     \n",
      "All:  0.962176782318 0.958857342316 33 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0201     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0202     \n",
      "All:  0.962142392886 0.958853903373 34 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0202     \n",
      "All:  0.96218881862 0.958872817561 35 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0198     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0203     \n",
      "All:  0.961972165193 0.958863360467 36 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0202     \n",
      "All:  0.96190080712 0.95888313439 37 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0197     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 3s - loss: 0.0202     \n",
      "All:  0.962353887897 0.958878835711 38 0\n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 0s - loss: 0.0197     \n",
      "Epoch 1/1\n",
      "4640/4640 [==============================] - 4s - loss: 0.0202     \n",
      "All:  0.96218881862 0.958874537032 39 0\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0699     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0474     \n",
      "All:  0.93350218531 0.945289745223 0 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0337     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0272     \n",
      "All:  0.945916552831 0.954352054669 1 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0278     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 4s - loss: 0.0247     \n",
      "All:  0.953230443773 0.957326835882 2 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0254     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0237     \n",
      "All:  0.956368740015 0.958136746255 3 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0241     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0229     \n",
      "All:  0.957641942967 0.958448446778 4 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0232     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 4s - loss: 0.0221     \n",
      "All:  0.958759295659 0.958536165777 5 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0226     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0214     \n",
      "All:  0.959700358711 0.958646027436 6 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0222     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0216     \n",
      "All:  0.959244730899 0.958818058872 7 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0220     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 4s - loss: 0.0214     \n",
      "All:  0.959192780812 0.958963689444 8 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0217     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 4s - loss: 0.0211     \n",
      "All:  0.95891514592 0.95899179359 9 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0217     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 4s - loss: 0.0214     \n",
      "All:  0.959518959227 0.959074402744 10 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0215     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 4s - loss: 0.0211     \n",
      "All:  0.959649260265 0.959136572521 11 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0213     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0210     \n",
      "All:  0.958983277182 0.959157011899 12 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0209     \n",
      "All:  0.959616897916 0.959157011899 13 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0211     \n",
      "All:  0.959564096188 0.959204703782 14 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0211     \n",
      "All:  0.959541101887 0.959221736598 15 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0210     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0211     \n",
      "All:  0.959741237468 0.959259208792 16 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0209     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0210     \n",
      "All:  0.959683325896 0.95924643418 17 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0210     \n",
      "All:  0.958745669407 0.959284758015 18 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0208     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0207     \n",
      "All:  0.95983151139 0.959299235908 19 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0207     \n",
      "All:  0.959973735399 0.959321378568 20 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0207     \n",
      "All:  0.959956702583 0.959319675286 21 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.959918378748 0.959304345753 22 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959667144721 0.959330746616 23 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.959775303099 0.959341817946 24 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0205     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.95957772244 0.959341817946 25 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.959757418643 0.959362257325 26 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0205     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0207     \n",
      "All:  0.959954999302 0.959373328655 27 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959781264584 0.959386954907 28 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.959919230389 0.959392064752 29 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.960220711222 0.95937929014 30 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0207     \n",
      "All:  0.96027351295 0.9594014328 31 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0203     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959977141962 0.959399729519 32 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0203     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.960090410184 0.959399729519 33 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0202     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0209     \n",
      "All:  0.960181535747 0.959409097567 34 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0202     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.959971180476 0.959405691004 35 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959888571322 0.959407394286 36 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.960098074951 0.959413355771 37 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0205     \n",
      "All:  0.960437027978 0.959405691004 38 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.960644828326 0.959415059053 39 1\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.1127     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0489     \n",
      "All:  0.856757407979 0.939524474846 0 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0514     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0288     \n",
      "All:  0.918679512826 0.952354743984 1 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0338     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0244     \n",
      "All:  0.939487205267 0.953076733729 2 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0271     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0232     \n",
      "All:  0.948820201152 0.953896664459 3 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0244     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0224     \n",
      "All:  0.952469152923 0.95460391972 4 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0233     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0220     \n",
      "All:  0.954036208696 0.954473042827 5 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0226     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0216     \n",
      "All:  0.955069356086 0.95448344364 6 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0223     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0209     \n",
      "All:  0.956493400684 0.954665457861 7 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0219     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0212     \n",
      "All:  0.956922434206 0.954756464972 8 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0218     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0208     \n",
      "All:  0.957816904094 0.954806735566 9 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0215     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0207     \n",
      "All:  0.958104659911 0.955089290977 10 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0214     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0210     \n",
      "All:  0.958364680227 0.954996550397 11 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0212     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0209     \n",
      "All:  0.958332611055 0.955101425258 12 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0208     \n",
      "All:  0.958919390235 0.955061555476 13 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.959232281349 0.955016485288 14 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0211     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.958670637466 0.955072823023 15 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0209     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959259150115 0.955152562587 16 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0209     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959019931424 0.955179431353 17 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.958925457376 0.955162096665 18 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0207     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0206     \n",
      "All:  0.958944525532 0.955145628712 19 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0208     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959302486834 0.955110959336 20 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0208     \n",
      "All:  0.959529571244 0.955121360149 21 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959093603847 0.955179431353 22 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0206     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959331089069 0.955195032572 23 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0205     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0202     \n",
      "All:  0.959389160273 0.955167297071 24 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959351890694 0.95520456665 25 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0204     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0203     \n",
      "All:  0.959194145036 0.955233168885 26 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0203     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959612777745 0.955241836229 27 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0202     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0202     \n",
      "All:  0.959566840822 0.955266538159 28 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0201     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959486234524 0.955265238057 29 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0201     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959081469565 0.955258304182 30 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0201     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0203     \n",
      "All:  0.959224480739 0.955284306214 31 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0201     \n",
      "All:  0.959243548896 0.955278239073 32 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0200     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0204     \n",
      "All:  0.959356224366 0.955286039683 33 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0199     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0202     \n",
      "All:  0.959404761492 0.955313775183 34 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0199     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959057201003 0.955314641917 35 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0198     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0205     \n",
      "All:  0.959110938535 0.955322442527 36 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0197     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0203     \n",
      "All:  0.959186344426 0.955313775183 37 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0197     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 3s - loss: 0.0203     \n",
      "All:  0.959102271191 0.955329376402 38 2\n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 0s - loss: 0.0197     \n",
      "Epoch 1/1\n",
      "4641/4641 [==============================] - 2s - loss: 0.0203     \n",
      "All:  0.959051133862 0.955335443543 39 2\n"
     ]
    }
   ],
   "source": [
    "folds = 3\n",
    "batch_size_nn = 16\n",
    "batch_size_lstm = 16\n",
    "hidden = 50\n",
    "dropout_probability = 0.25\n",
    "\n",
    "n_epochs = 40\n",
    "epoch = 0\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "train_nn_aucs = np.zeros((folds,n_epochs))\n",
    "test_nn_aucs = np.zeros((folds,n_epochs))\n",
    "\n",
    "\n",
    "\n",
    "train_lstm_aucs = np.zeros((folds,n_epochs))\n",
    "test_lstm_aucs = np.zeros((folds,n_epochs))\n",
    "\n",
    "\n",
    "\n",
    "for i, (train_idx, test_idx) in enumerate(KFold(len(X_9),folds, shuffle=True)):\n",
    "    \n",
    "\n",
    "    \n",
    "    \n",
    "    # neural net\n",
    "    nn = Sequential()\n",
    "    nn.add(Embedding(input_dim = 21, input_length =9, output_dim= 32))\n",
    "    nn.add(Flatten())\n",
    "    nn.add(Dense(10, init='glorot_uniform', activation='sigmoid'))\n",
    "    nn.add(Dense(1, init='glorot_uniform', activation='sigmoid'))\n",
    "    \n",
    "    nn.compile(optimizer = 'adam', loss='mean_squared_error')\n",
    "    \n",
    "    \n",
    "    # lstm \n",
    "    sequence = Input( shape= (9, ),dtype='int32')\n",
    "    embedded = Embedding(input_dim = 21, input_length = 9, output_dim= 32)(sequence)\n",
    "    forwards = LSTM(hidden)(embedded)\n",
    "    backwards = LSTM(hidden, go_backwards=True)(embedded)\n",
    "\n",
    "    merged = merge([forwards, backwards], mode = 'concat', concat_axis=-1)\n",
    "    after_dp = Dropout(dropout_probability)(merged)\n",
    "    output = Dense(1, activation = 'sigmoid')(after_dp)\n",
    "    lstm = Model(input = sequence, output = output)\n",
    "    adam = Adam(lr=0.01)\n",
    "    lstm.compile(optimizer = adam, loss='mean_squared_error')\n",
    "    \n",
    "    \n",
    "    # index sets\n",
    "    \n",
    "    nine_train_idx = np.array([i for i in train_idx if (np.count_nonzero(X[i])==9)])\n",
    "    non_nine_train_idx = np.array([i for i in train_idx if (np.count_nonzero(X[i])!=9)])\n",
    "    nine_test_idx = np.array([i for i in test_idx if (np.count_nonzero(X[i])==9)])\n",
    "    non_nine_test_idx = np.array([i for i in test_idx if (np.count_nonzero(X[i])!=9)])\n",
    "    \n",
    "  \n",
    "    \n",
    "    for epoch in range(n_epochs):\n",
    "        \n",
    "        \n",
    "        \n",
    "        \n",
    "        # nn\n",
    "        nn.fit(X_9_shortened[train_idx],y_9[train_idx], batch_size = batch_size_nn, nb_epoch=1)\n",
    "   \n",
    "\n",
    "        train_nn_auc = roc_auc_score(measured_affinity_less_than(y_9[train_idx],500), nn.predict(X_9_shortened[train_idx]))     \n",
    "        test_nn_auc = roc_auc_score(measured_affinity_less_than(y_9[test_idx],500), nn.predict(X_9_shortened[test_idx]))\n",
    "        \n",
    "        \n",
    "        \n",
    "        train_nn_aucs[i][epoch]=train_nn_auc\n",
    "        test_nn_aucs[i][epoch]=test_nn_auc\n",
    "        \n",
    "        \n",
    "        \n",
    "        #lstm \n",
    "        adam.lr.set_value(0.01*(epoch+1)**(-2))\n",
    "        lstm.fit(X_9_shortened[train_idx],y_9[train_idx], batch_size = batch_size_lstm, nb_epoch=1)\n",
    "    \n",
    "    \n",
    "        train_lstm_auc = roc_auc_score(measured_affinity_less_than(y_9[train_idx],500), lstm.predict(X_9_shortened[train_idx]))     \n",
    "        test_lstm_auc = roc_auc_score(measured_affinity_less_than(y_9[test_idx],500), lstm.predict(X_9_shortened[test_idx]))\n",
    "        \n",
    "        train_lstm_aucs[i][epoch]=train_lstm_auc\n",
    "        test_lstm_aucs[i][epoch]=test_lstm_auc\n",
    "\n",
    "        \n",
    "        print(\"All: \", test_nn_auc, test_lstm_auc, epoch, i)\n",
    "  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "train_nn_aucs_mean = np.mean(train_nn_aucs, axis=0)\n",
    "test_nn_aucs_mean = np.mean(test_nn_aucs, axis=0)\n",
    "train_lstm_aucs_mean=np.mean(train_lstm_aucs, axis=0)\n",
    "test_lstm_aucs_mean=np.mean(test_lstm_aucs, axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x135d8a978>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhEAAAFoCAYAAADpW6mGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x118a612e8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot( np.arange(0,40,1), test_nn_aucs_mean[0:40],color='orange',marker='*', linestyle='-', label =\"test NN \")\n",
    "plt.plot( np.arange(0,40,1), test_lstm_aucs_mean[0:40],color='g',marker='*', linestyle='-', label =\"test LSTM (with LR) \")\n",
    "\n",
    "plt.legend(loc=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}