MTL-Tensorflow.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(8000, 100)\n",
      "(1000, 100)\n",
      "(1000, 100)\n",
      "(8000, 1)\n",
      "(8000, 1)\n",
      "(1000, 1)\n",
      "(1000, 1)\n",
      "(1000, 1)\n",
      "(1000, 1)\n",
      "(8000, 2)\n",
      "(1000, 2)\n",
      "(1000, 2)\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "#import torch\n",
    "#import torch.nn as nn\n",
    "#import torchvision\n",
    "#import torchvision.transforms as transforms\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.gridspec as gridspec\n",
    "import os\n",
    "#from torch.autograd import Variable\n",
    "import pandas as pd\n",
    "import math\n",
    "import sklearn.preprocessing as sk\n",
    "#from tensorboardX import SummaryWriter\n",
    "#import seaborn as sns\n",
    "#from sklearn.model_selection import KFold\n",
    "from sklearn import metrics\n",
    "#from sklearn.feature_selection import VarianceThreshold\n",
    "#from sklearn.linear_model import Ridge\n",
    "#from sklearn.linear_model import RidgeCV\n",
    "#from sklearn.model_selection import train_test_split\n",
    "import random\n",
    "\n",
    "seed = 42\n",
    "random.seed(seed)\n",
    "\n",
    "N = 10000\n",
    "M = 100\n",
    "c = 0.5\n",
    "p = 0.9\n",
    "k = np.random.randn(M)\n",
    "u1 = np.random.randn(M)\n",
    "u1 -= u1.dot(k) * k / np.linalg.norm(k)**2\n",
    "u1 /= np.linalg.norm(u1) \n",
    "k /= np.linalg.norm(k) \n",
    "u2 = k\n",
    "w1 = c*u1\n",
    "w2 = c*(p*u1+np.sqrt((1-p**2))*u2)\n",
    "X = np.random.normal(0, 1, (N, M))\n",
    "eps1 = np.random.normal(0, 0.01)\n",
    "eps2 = np.random.normal(0, 0.01)\n",
    "Y1 = np.matmul(X, w1) + np.sin(np.matmul(X, w1))+eps1\n",
    "Y2 = np.matmul(X, w2) + np.sin(np.matmul(X, w2))+eps2\n",
    "split = list(np.random.permutation(N))\n",
    "\n",
    "X_train = X[split[0:8000],:]\n",
    "Y1_train = Y1[split[0:8000]].reshape(-1,1)\n",
    "Y2_train = Y2[split[0:8000]].reshape(-1,1)\n",
    "X_valid = X[8000:9000,:]\n",
    "Y1_valid = Y1[8000:9000].reshape(-1,1)\n",
    "Y2_valid = Y2[8000:9000].reshape(-1,1)\n",
    "X_test = X[9000:10000,:]\n",
    "Y1_test = Y1[9000:10000].reshape(-1,1)\n",
    "Y2_test = Y2[9000:10000].reshape(-1,1)\n",
    "print(X_train.shape)\n",
    "print(X_valid.shape)\n",
    "print(X_test.shape)\n",
    "print(Y1_train.shape)\n",
    "print(Y2_train.shape)\n",
    "print(Y1_valid.shape)\n",
    "print(Y2_valid.shape)\n",
    "print(Y1_test.shape)\n",
    "print(Y2_test.shape)\n",
    "Y_train = np.concatenate((Y1_train.reshape(-1,1), Y2_train.reshape(-1,1)), axis=1)\n",
    "Y_valid = np.concatenate((Y1_valid.reshape(-1,1), Y2_valid.reshape(-1,1)), axis=1)\n",
    "Y_test = np.concatenate((Y1_test.reshape(-1,1), Y2_test.reshape(-1,1)), axis=1)\n",
    "print(Y_train.shape)\n",
    "print(Y_valid.shape)\n",
    "print(Y_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 0, Train Cost: 1.930 \n",
      "Epoch: 0, Dev Cost: 1.204 \n",
      "Epoch: 1, Train Cost: 1.065 \n",
      "Epoch: 1, Dev Cost: 1.019 \n",
      "Epoch: 2, Train Cost: 0.961 \n",
      "Epoch: 2, Dev Cost: 0.938 \n",
      "Epoch: 3, Train Cost: 0.929 \n",
      "Epoch: 3, Dev Cost: 0.918 \n",
      "Epoch: 4, Train Cost: 0.914 \n",
      "Epoch: 4, Dev Cost: 0.901 \n",
      "Epoch: 5, Train Cost: 0.900 \n",
      "Epoch: 5, Dev Cost: 0.897 \n",
      "Epoch: 6, Train Cost: 0.883 \n",
      "Epoch: 6, Dev Cost: 0.883 \n",
      "Epoch: 7, Train Cost: 0.864 \n",
      "Epoch: 7, Dev Cost: 0.845 \n",
      "Epoch: 8, Train Cost: 0.807 \n",
      "Epoch: 8, Dev Cost: 0.774 \n",
      "Epoch: 9, Train Cost: 0.722 \n",
      "Epoch: 9, Dev Cost: 0.666 \n",
      "Epoch: 10, Train Cost: 0.619 \n",
      "Epoch: 10, Dev Cost: 0.567 \n",
      "Epoch: 11, Train Cost: 0.548 \n",
      "Epoch: 11, Dev Cost: 0.507 \n",
      "Epoch: 12, Train Cost: 0.482 \n",
      "Epoch: 12, Dev Cost: 0.452 \n",
      "Epoch: 13, Train Cost: 0.442 \n",
      "Epoch: 13, Dev Cost: 0.439 \n",
      "Epoch: 14, Train Cost: 0.429 \n",
      "Epoch: 14, Dev Cost: 0.428 \n",
      "Epoch: 15, Train Cost: 0.396 \n",
      "Epoch: 15, Dev Cost: 0.373 \n",
      "Epoch: 16, Train Cost: 0.378 \n",
      "Epoch: 16, Dev Cost: 0.370 \n",
      "Epoch: 17, Train Cost: 0.356 \n",
      "Epoch: 17, Dev Cost: 0.365 \n",
      "Epoch: 18, Train Cost: 0.344 \n",
      "Epoch: 18, Dev Cost: 0.338 \n",
      "Epoch: 19, Train Cost: 0.333 \n",
      "Epoch: 19, Dev Cost: 0.343 \n",
      "Epoch: 20, Train Cost: 0.327 \n",
      "Epoch: 20, Dev Cost: 0.315 \n",
      "Epoch: 21, Train Cost: 0.316 \n",
      "Epoch: 21, Dev Cost: 0.328 \n",
      "Epoch: 22, Train Cost: 0.304 \n",
      "Epoch: 22, Dev Cost: 0.328 \n",
      "Epoch: 23, Train Cost: 0.306 \n",
      "Epoch: 23, Dev Cost: 0.303 \n",
      "Epoch: 24, Train Cost: 0.299 \n",
      "Epoch: 24, Dev Cost: 0.278 \n",
      "Epoch: 25, Train Cost: 0.283 \n",
      "Epoch: 25, Dev Cost: 0.300 \n",
      "Epoch: 26, Train Cost: 0.288 \n",
      "Epoch: 26, Dev Cost: 0.274 \n",
      "Epoch: 27, Train Cost: 0.277 \n",
      "Epoch: 27, Dev Cost: 0.294 \n",
      "Epoch: 28, Train Cost: 0.273 \n",
      "Epoch: 28, Dev Cost: 0.275 \n",
      "Epoch: 29, Train Cost: 0.266 \n",
      "Epoch: 29, Dev Cost: 0.268 \n",
      "Epoch: 30, Train Cost: 0.261 \n",
      "Epoch: 30, Dev Cost: 0.286 \n",
      "Epoch: 31, Train Cost: 0.256 \n",
      "Epoch: 31, Dev Cost: 0.259 \n",
      "Epoch: 32, Train Cost: 0.253 \n",
      "Epoch: 32, Dev Cost: 0.249 \n",
      "Epoch: 33, Train Cost: 0.250 \n",
      "Epoch: 33, Dev Cost: 0.251 \n",
      "Epoch: 34, Train Cost: 0.245 \n",
      "Epoch: 34, Dev Cost: 0.246 \n",
      "Epoch: 35, Train Cost: 0.243 \n",
      "Epoch: 35, Dev Cost: 0.232 \n",
      "Epoch: 36, Train Cost: 0.238 \n",
      "Epoch: 36, Dev Cost: 0.239 \n",
      "Epoch: 37, Train Cost: 0.232 \n",
      "Epoch: 37, Dev Cost: 0.225 \n",
      "Epoch: 38, Train Cost: 0.228 \n",
      "Epoch: 38, Dev Cost: 0.229 \n",
      "Epoch: 39, Train Cost: 0.228 \n",
      "Epoch: 39, Dev Cost: 0.231 \n",
      "Epoch: 40, Train Cost: 0.225 \n",
      "Epoch: 40, Dev Cost: 0.231 \n",
      "Epoch: 41, Train Cost: 0.216 \n",
      "Epoch: 41, Dev Cost: 0.210 \n",
      "Epoch: 42, Train Cost: 0.218 \n",
      "Epoch: 42, Dev Cost: 0.210 \n",
      "Epoch: 43, Train Cost: 0.222 \n",
      "Epoch: 43, Dev Cost: 0.231 \n",
      "Epoch: 44, Train Cost: 0.212 \n",
      "Epoch: 44, Dev Cost: 0.217 \n",
      "Epoch: 45, Train Cost: 0.207 \n",
      "Epoch: 45, Dev Cost: 0.222 \n",
      "Epoch: 46, Train Cost: 0.211 \n",
      "Epoch: 46, Dev Cost: 0.207 \n",
      "Epoch: 47, Train Cost: 0.206 \n",
      "Epoch: 47, Dev Cost: 0.205 \n",
      "Epoch: 48, Train Cost: 0.199 \n",
      "Epoch: 48, Dev Cost: 0.206 \n",
      "Epoch: 49, Train Cost: 0.200 \n",
      "Epoch: 49, Dev Cost: 0.201 \n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEKCAYAAAAB0GKPAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n28XRCtZh+wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4f138c3f90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEKCAYAAAD9xUlFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4f13882fd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEKCAYAAAAB0GKPAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4f12e1cf10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learning_rate = 0.001\n",
    "mb_size = 100\n",
    "n_shared = 64\n",
    "n_hidden1 = 32\n",
    "n_hidden2 = 16\n",
    "num_epochs = 50\n",
    "#keep_prob = 1\n",
    "n_output = 1\n",
    "\n",
    "def PlaceHolder(dims):\n",
    "    PH = tf.placeholder(tf.float32, [None, dims])\n",
    "    return PH \n",
    "\n",
    "def Shared_layer(dat, keep_prob):\n",
    "\n",
    "    H1 = tf.layers.dense(inputs=dat, units=n_shared, activation=tf.nn.relu)\n",
    "    Drop_H1 = tf.nn.dropout(H1, keep_prob)\n",
    "    #bn_Drop_H1 = tf.contrib.layers.batch_norm(Drop_H1, center =True, scale = True)    \n",
    "\n",
    "    return Drop_H1\n",
    "\n",
    "def Tower_layer1(dat, keep_prob):\n",
    "\n",
    "    H1_1 = tf.layers.dense(inputs=dat, units=n_hidden1, activation=tf.nn.relu)\n",
    "    Drop_H1_1 = tf.nn.dropout(H1_1, keep_prob)\n",
    "    #bn_Drop_H1 = tf.contrib.layers.batch_norm(Drop_H1, center =True, scale = True)\n",
    "\n",
    "    H2_1 = tf.layers.dense(inputs=Drop_H1_1, units=n_hidden2, activation=tf.nn.relu)\n",
    "    Drop_H2_1 = tf.nn.dropout(H2_1, keep_prob)\n",
    "    #bn_Drop_H2 = tf.contrib.layers.batch_norm(Drop_H2, center =True, scale = True)\n",
    "\n",
    "    H3_1 = tf.layers.dense(inputs=Drop_H2_1, units=n_output)\n",
    "\n",
    "    return H3_1\n",
    "\n",
    "def Tower_layer2(dat, keep_prob):\n",
    "\n",
    "    H1_2 = tf.layers.dense(inputs=dat, units=n_hidden1, activation=tf.nn.relu)\n",
    "    Drop_H1_2 = tf.nn.dropout(H1_2, keep_prob)\n",
    "    #bn_Drop_H1 = tf.contrib.layers.batch_norm(Drop_H1, center =True, scale = True)\n",
    "\n",
    "    H2_2 = tf.layers.dense(inputs=Drop_H1_2, units=n_hidden2, activation=tf.nn.relu)\n",
    "    Drop_H2_2 = tf.nn.dropout(H2_2, keep_prob)\n",
    "    #bn_Drop_H2 = tf.contrib.layers.batch_norm(Drop_H2, center =True, scale = True)\n",
    "\n",
    "    H3_2 = tf.layers.dense(inputs=Drop_H2_2, units=n_output)\n",
    "\n",
    "    return H3_2\n",
    "\n",
    "def compute_cost(Y_pred, Y_true):\n",
    "\n",
    "    Logits = Y_pred\n",
    "    Labels = Y_true \n",
    "    #Labels = tf.reshape(Labels, [-1,1])\n",
    "    Loss = tf.losses.mean_squared_error(Labels, Logits)\n",
    "    cost = tf.reduce_mean(Loss)       \n",
    "    return cost\n",
    "\n",
    "def random_mini_batches(XE, R1E, R2E, mini_batch_size = 10, seed = 42): \n",
    "    # Creating the mini-batches\n",
    "    np.random.seed(seed)            \n",
    "    m = XE.shape[0]                  \n",
    "    mini_batches = []\n",
    "    permutation = list(np.random.permutation(m))\n",
    "    shuffled_XE = XE[permutation,:]\n",
    "    shuffled_X1R = R1E[permutation]\n",
    "    shuffled_X2R = R2E[permutation]\n",
    "    num_complete_minibatches = math.floor(m/mini_batch_size)\n",
    "    for k in range(0, int(num_complete_minibatches)):\n",
    "        mini_batch_XE = shuffled_XE[k * mini_batch_size : (k+1) * mini_batch_size, :]\n",
    "        mini_batch_X1R = shuffled_X1R[k * mini_batch_size : (k+1) * mini_batch_size]\n",
    "        mini_batch_X2R = shuffled_X2R[k * mini_batch_size : (k+1) * mini_batch_size]\n",
    "        mini_batch = (mini_batch_XE, mini_batch_X1R, mini_batch_X2R)\n",
    "        mini_batches.append(mini_batch)\n",
    "    Lower = int(num_complete_minibatches * mini_batch_size)\n",
    "    Upper = int(m - (mini_batch_size * math.floor(m/mini_batch_size)))\n",
    "    if m % mini_batch_size != 0:\n",
    "        mini_batch_XE = shuffled_XE[Lower : Lower + Upper, :]\n",
    "        mini_batch_X1R = shuffled_X1R[Lower : Lower + Upper]\n",
    "        mini_batch_X2R = shuffled_X2R[Lower : Lower + Upper]\n",
    "        mini_batch = (mini_batch_XE, mini_batch_X1R, mini_batch_X2R)\n",
    "        mini_batches.append(mini_batch)\n",
    "    \n",
    "    return mini_batches\n",
    "\n",
    "seed = 42 \n",
    "tf.set_random_seed(seed)    \n",
    "(n_samp, n_gene) = X_train.shape\n",
    "(n_samp, n_out) = Y_train.shape\n",
    "cost1tr = []\n",
    "cost2tr = []\n",
    "cost1D = []\n",
    "cost2D = []\n",
    "cost1ts = []\n",
    "cost2ts = []\n",
    "costtr = []\n",
    "costD = []\n",
    "costts = []\n",
    "\n",
    "tf.reset_default_graph()\n",
    "\n",
    "X = PlaceHolder(n_gene)\n",
    "Y1 = PlaceHolder(n_output)\n",
    "Y2 = PlaceHolder(n_output)\n",
    "kp = tf.placeholder(tf.float32)\n",
    "\n",
    "H_shared = Shared_layer(X, kp)\n",
    "H1 = Tower_layer1(H_shared, kp)\n",
    "H2 = Tower_layer2(H_shared, kp)\n",
    "        \n",
    "cost1 = compute_cost(H1, Y1)\n",
    "cost2 = compute_cost(H2, Y2)\n",
    "\n",
    "cost = tf.divide(tf.add(cost1, cost2), 2)\n",
    "\n",
    "optimizer = tf.train.AdamOptimizer(learning_rate = learning_rate).minimize(cost)\n",
    "    \n",
    "init = tf.global_variables_initializer()\n",
    "\n",
    "with tf.Session() as sess:       \n",
    "    sess.run(init)\n",
    "    for epoch in range(num_epochs):\n",
    "        epoch_cost = 0  \n",
    "        epoch_cost1 = 0\n",
    "        epoch_cost2 = 0\n",
    "        \n",
    "        num_minibatches = int(n_samp / mb_size) \n",
    "        minibatches = random_mini_batches(X_train, Y1_train, Y2_train, mb_size, seed)\n",
    "\n",
    "        for minibatch in minibatches:\n",
    "            (mb_X, mb_Y1, mb_Y2) = minibatch\n",
    "            _ , minibatch_cost= sess.run([optimizer, cost], feed_dict={X: mb_X, Y1: mb_Y1, Y2: mb_Y2, kp: 0.5})\n",
    "            epoch_cost = epoch_cost + (minibatch_cost / num_minibatches)\n",
    "            \n",
    "            mb_cost1 = sess.run(cost1, feed_dict={X: mb_X, Y1: mb_Y1, Y2: mb_Y2, kp: 0.5})\n",
    "            epoch_cost1 = epoch_cost1 + (mb_cost1 / num_minibatches)\n",
    "            \n",
    "            mb_cost2 = sess.run(cost2, feed_dict={X: mb_X, Y1: mb_Y1, Y2: mb_Y2, kp: 0.5})\n",
    "            epoch_cost2 = epoch_cost2 + (mb_cost2 / num_minibatches)\n",
    "            \n",
    "        if epoch % 1 == 0:\n",
    "            print('Epoch: %d, Train Cost: %5.3f ' % (epoch, epoch_cost))\n",
    "            dv_cost = sess.run(cost, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, kp: 0.5})\n",
    "            dv_cost1 = sess.run(cost1, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, kp: 0.5})\n",
    "            cost1D.append(dv_cost1)\n",
    "            dv_cost2 = sess.run(cost2, feed_dict={X: X_valid, Y1: Y1_valid, Y2: Y2_valid, kp: 0.5})\n",
    "            cost2D.append(dv_cost2)\n",
    "            #dv_cost = (dv_cost1+dv_cost2)/2\n",
    "\n",
    "            print('Epoch: %d, Dev Cost: %5.3f ' % (epoch, dv_cost))\n",
    "        costtr.append(np.mean(epoch_cost))\n",
    "        cost1tr.append(np.mean(epoch_cost1))\n",
    "        cost2tr.append(np.mean(epoch_cost2))\n",
    "        costD.append(dv_cost)\n",
    "\n",
    "plt.plot(np.squeeze(costtr),'-r',np.squeeze(costD), '-b')\n",
    "plt.ylabel('total cost')\n",
    "plt.xlabel('iterations')\n",
    "plt.show()\n",
    "\n",
    "plt.plot(np.squeeze(cost1tr), '-r', np.squeeze(cost1D), '-b')\n",
    "plt.ylabel('task 1 cost')\n",
    "plt.xlabel('iterations (per tens)')\n",
    "plt.show() \n",
    "\n",
    "plt.plot(np.squeeze(cost2tr),'-r', np.squeeze(cost2D),'-b')\n",
    "plt.ylabel('task 2 cost')\n",
    "plt.xlabel('iterations (per tens)')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}