result_rsenet.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import RSE Net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Anaconda3\\lib\\site-packages\\h5py\\__init__.py:34: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import rse_net  # rse_net.py"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load Data Set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from keras.models import Sequential\n",
    "from keras.models import Model, Input\n",
    "from keras.layers import Dense, Activation, Dropout, BatchNormalization, Average\n",
    "from keras.optimizers import SGD\n",
    "from keras.callbacks import EarlyStopping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(450, 30)\n",
      "(450,)\n"
     ]
    }
   ],
   "source": [
    "X = np.load('features_AC.npy')\n",
    "y = np.load('y_response.npy')\n",
    "print(X.shape)\n",
    "print(y.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelBinarizer \n",
    "label_binarizer = LabelBinarizer()\n",
    "label_binarizer.fit([1,2,3]) # need to be global or remembered to use it later\n",
    "\n",
    "def one_hot_encode(x):\n",
    "    return label_binarizer.transform(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(450, 3)\n"
     ]
    }
   ],
   "source": [
    "y_ohe = one_hot_encode(y)\n",
    "print(y_ohe.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Features in Random Subspace"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "rsf = rse_net.rs_features(X, n_classifier = 7, f_lb = 20, f_ub = 30) # . for reminder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(450, 30)\n",
      "(450, 21)\n",
      "(450, 25)\n",
      "(450, 29)\n",
      "(450, 23)\n",
      "(450, 26)\n",
      "(450, 22)\n"
     ]
    }
   ],
   "source": [
    "for i in range(len(rsf)):\n",
    "    print(rsf[i].shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Ensemble Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "models = []\n",
    "input_layers = []\n",
    "for i in range(len(rsf)):\n",
    "    md_n, il_n = rse_net.classifier(rsf[i].shape[1]) # . for reminder\n",
    "    models.append(md_n)\n",
    "    input_layers.append(il_n)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model_rse = rse_net.ensemble(models, input_layers) # . for reminder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 30)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_2 (InputLayer)            (None, 21)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_3 (InputLayer)            (None, 25)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_4 (InputLayer)            (None, 29)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_5 (InputLayer)            (None, 23)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_6 (InputLayer)            (None, 26)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_7 (InputLayer)            (None, 22)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 32)           992         input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_5 (Dense)                 (None, 32)           704         input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_9 (Dense)                 (None, 32)           832         input_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_13 (Dense)                (None, 32)           960         input_4[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_17 (Dense)                (None, 32)           768         input_5[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_21 (Dense)                (None, 32)           864         input_6[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_25 (Dense)                (None, 32)           736         input_7[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 32)           0           dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_4 (Dropout)             (None, 32)           0           dense_5[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_7 (Dropout)             (None, 32)           0           dense_9[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_10 (Dropout)            (None, 32)           0           dense_13[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_13 (Dropout)            (None, 32)           0           dense_17[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_16 (Dropout)            (None, 32)           0           dense_21[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_19 (Dropout)            (None, 32)           0           dense_25[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 64)           2112        dropout_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_6 (Dense)                 (None, 64)           2112        dropout_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_10 (Dense)                (None, 64)           2112        dropout_7[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_14 (Dense)                (None, 64)           2112        dropout_10[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_18 (Dense)                (None, 64)           2112        dropout_13[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_22 (Dense)                (None, 64)           2112        dropout_16[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_26 (Dense)                (None, 64)           2112        dropout_19[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dropout_2 (Dropout)             (None, 64)           0           dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_5 (Dropout)             (None, 64)           0           dense_6[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_8 (Dropout)             (None, 64)           0           dense_10[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_11 (Dropout)            (None, 64)           0           dense_14[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_14 (Dropout)            (None, 64)           0           dense_18[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_17 (Dropout)            (None, 64)           0           dense_22[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_20 (Dropout)            (None, 64)           0           dense_26[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 128)          8320        dropout_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_7 (Dense)                 (None, 128)          8320        dropout_5[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_11 (Dense)                (None, 128)          8320        dropout_8[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_15 (Dense)                (None, 128)          8320        dropout_11[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_19 (Dense)                (None, 128)          8320        dropout_14[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_23 (Dense)                (None, 128)          8320        dropout_17[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_27 (Dense)                (None, 128)          8320        dropout_20[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dropout_3 (Dropout)             (None, 128)          0           dense_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_6 (Dropout)             (None, 128)          0           dense_7[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_9 (Dropout)             (None, 128)          0           dense_11[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_12 (Dropout)            (None, 128)          0           dense_15[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_15 (Dropout)            (None, 128)          0           dense_19[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_18 (Dropout)            (None, 128)          0           dense_23[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_21 (Dropout)            (None, 128)          0           dense_27[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 3)            387         dropout_3[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_8 (Dense)                 (None, 3)            387         dropout_6[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_12 (Dense)                (None, 3)            387         dropout_9[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_16 (Dense)                (None, 3)            387         dropout_12[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_20 (Dense)                (None, 3)            387         dropout_15[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_24 (Dense)                (None, 3)            387         dropout_18[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_28 (Dense)                (None, 3)            387         dropout_21[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "average_1 (Average)             (None, 3)            0           dense_4[0][0]                    \n",
      "                                                                 dense_8[0][0]                    \n",
      "                                                                 dense_12[0][0]                   \n",
      "                                                                 dense_16[0][0]                   \n",
      "                                                                 dense_20[0][0]                   \n",
      "                                                                 dense_24[0][0]                   \n",
      "                                                                 dense_28[0][0]                   \n",
      "==================================================================================================\n",
      "Total params: 81,589\n",
      "Trainable params: 81,589\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_rse.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from keras.optimizers import Adam\n",
    "adam = Adam(lr=0.02, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)\n",
    "# kept default params"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model_rse.compile(optimizer=adam,\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train Ensemble Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "mhist = model_rse.fit(rsf, y_ohe, batch_size = 512, validation_split = 0.1,\n",
    "             epochs=100, verbose = 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8222222328186035"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.max(mhist.history['val_acc'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### 10-fold CV"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "from rse_net import cross_validation\n",
    "v_acc_a = rse_net.cross_validation(rsf, y_ohe, k = 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.9333333373069763, 0.8666666746139526, 0.9111111164093018, 0.8444444537162781, 0.8222222328186035, 0.7555555701255798, 0.8888888955116272, 0.8666666746139526, 0.8666666746139526, 0.8444444537162781]\n"
     ]
    }
   ],
   "source": [
    "print(v_acc_a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy:  0.8600000083446503  +/-  0.09333332777023315\n"
     ]
    }
   ],
   "source": [
    "print('Accuracy: ', np.mean(v_acc_a), ' +/- ', 2*np.std(v_acc_a))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}