Tensorflow-task.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Going deeper with Tensorflow\n",
    "\n",
    "In this video, we're going to study the tools you'll use to build deep learning models. Namely, [Tensorflow](https://www.tensorflow.org/).\n",
    "\n",
    "If you're running this notebook outside the course environment, you'll need to install tensorflow:\n",
    "* `pip install tensorflow` should install cpu-only TF on Linux & Mac OS\n",
    "* If you want GPU support from offset, see [TF install page](https://www.tensorflow.org/install/)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import sys\n",
    "sys.path.append(\"..\")\n",
    "import grading"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visualization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plase note that if you are running on the Coursera platform, you won't be able to access the tensorboard instance due to the network setup there. If you run the notebook locally, you should be able to access TensorBoard on http://127.0.0.1:7007/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/bin/sh: 1: killall: not found\r\n"
     ]
    }
   ],
   "source": [
    "! killall tensorboard\n",
    "import os\n",
    "os.system(\"tensorboard --logdir=/tmp/tboard --port=7007 &\");"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "s = tf.InteractiveSession()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Warming up\n",
    "For starters, let's implement a python function that computes the sum of squares of numbers from 0 to N-1."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "def sum_sin(N):\n",
    "    return np.sum(np.arange(N)**2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 396 ms, sys: 316 ms, total: 712 ms\n",
      "Wall time: 1.1 s\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "662921401752298880"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "sum_sin(10**8)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tensoflow teaser\n",
    "\n",
    "Doing the very same thing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# An integer parameter\n",
    "N = tf.placeholder('int64', name=\"input_to_your_function\")\n",
    "\n",
    "# A recipe on how to produce the same result\n",
    "result = tf.reduce_sum(tf.range(N)**2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor 'Sum:0' shape=() dtype=int64>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 468 ms, sys: 140 ms, total: 608 ms\n",
      "Wall time: 491 ms\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "662921401752298880"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "result.eval({N: 10**8})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "writer = tf.summary.FileWriter(\"/tmp/tboard\", graph=s.graph)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# How does it work?\n",
    "1. Define placeholders where you'll send inputs\n",
    "2. Make symbolic graph: a recipe for mathematical transformation of those placeholders\n",
    "3. Compute outputs of your graph with particular values for each placeholder\n",
    "  * `output.eval({placeholder:value})`\n",
    "  * `s.run(output, {placeholder:value})`\n",
    "\n",
    "So far there are two main entities: \"placeholder\" and \"transformation\"\n",
    "* Both can be numbers, vectors, matrices, tensors, etc.\n",
    "* Both can be int32/64, floats, booleans (uint8) of various size.\n",
    "\n",
    "* You can define new transformations as an arbitrary operation on placeholders and other transformations\n",
    " * `tf.reduce_sum(tf.arange(N)**2)` are 3 sequential transformations of placeholder `N`\n",
    " * There's a tensorflow symbolic version for every numpy function\n",
    "   * `a+b, a/b, a**b, ...` behave just like in numpy\n",
    "   * `np.mean` -> `tf.reduce_mean`\n",
    "   * `np.arange` -> `tf.range`\n",
    "   * `np.cumsum` -> `tf.cumsum`\n",
    "   * If if you can't find the op you need, see the [docs](https://www.tensorflow.org/api_docs/python).\n",
    "   \n",
    "`tf.contrib` has many high-level features, may be worth a look."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"Placeholders_examples\"):\n",
    "    # Default placeholder that can be arbitrary float32\n",
    "    # scalar, vertor, matrix, etc.\n",
    "    arbitrary_input = tf.placeholder('float32')\n",
    "\n",
    "    # Input vector of arbitrary length\n",
    "    input_vector = tf.placeholder('float32', shape=(None,))\n",
    "\n",
    "    # Input vector that _must_ have 10 elements and integer type\n",
    "    fixed_vector = tf.placeholder('int32', shape=(10,))\n",
    "\n",
    "    # Matrix of arbitrary n_rows and 15 columns\n",
    "    # (e.g. a minibatch your data table)\n",
    "    input_matrix = tf.placeholder('float32', shape=(None, 15))\n",
    "    \n",
    "    # You can generally use None whenever you don't need a specific shape\n",
    "    input1 = tf.placeholder('float64', shape=(None, 100, None))\n",
    "    input2 = tf.placeholder('int32', shape=(None, None, 3, 224, 224))\n",
    "\n",
    "    # elementwise multiplication\n",
    "    double_the_vector = input_vector*2\n",
    "\n",
    "    # elementwise cosine\n",
    "    elementwise_cosine = tf.cos(input_vector)\n",
    "\n",
    "    # difference between squared vector and vector itself plus one\n",
    "    vector_squares = input_vector**2 - input_vector + 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "my_vector =  tf.placeholder('float32', shape=(None,), name=\"VECTOR_1\")\n",
    "my_vector2 = tf.placeholder('float32', shape=(None,))\n",
    "my_transformation = my_vector * my_vector2 / (tf.sin(my_vector) + 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"truediv:0\", shape=(?,), dtype=float32)\n"
     ]
    }
   ],
   "source": [
    "print(my_transformation)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.  1.  2.  3.  4.]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([ 0.        ,  1.62913239,  2.09501147,  2.62899613,  0.        ], dtype=float32)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dummy = np.arange(5).astype('float32')\n",
    "print(dummy)\n",
    "my_transformation.eval({my_vector:dummy, my_vector2:dummy[::-1]})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "writer.add_graph(my_transformation.graph)\n",
    "writer.flush()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "TensorBoard allows writing scalars, images, audio, histogram. You can read more on tensorboard usage [here](https://www.tensorflow.org/get_started/graph_viz)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Summary\n",
    "* Tensorflow is based on computation graphs\n",
    "* The graphs consist of placehlders and transformations"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Mean squared error\n",
    "\n",
    "Your assignment is to implement mean squared error in tensorflow."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"MSE\"):\n",
    "    y_true = tf.placeholder(\"float32\", shape=(None,), name=\"y_true\")\n",
    "    y_predicted = tf.placeholder(\"float32\", shape=(None,), name=\"y_predicted\")\n",
    "    mse = tf.reduce_mean((y_true-y_predicted)**2)\n",
    "    \n",
    "def compute_mse(vector1, vector2):\n",
    "    return mse.eval({y_true: vector1, y_predicted: vector2})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "writer.add_graph(mse.graph)\n",
    "writer.flush()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tests and result submission. Please use the credentials obtained from the Coursera assignment page."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import submit\n",
    "email = 'liberty.sky87@gmail.com'\n",
    "token = 'o3u2oXpPixnBwyCz'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "#submit.submit_mse(compute_mse, email, token)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Variables\n",
    "\n",
    "The inputs and transformations have no value outside function call. This isn't too comfortable if you want your model to have parameters (e.g. network weights) that are always present, but can change their value over time.\n",
    "\n",
    "Tensorflow solves this with `tf.Variable` objects.\n",
    "* You can assign variable a value at any time in your graph\n",
    "* Unlike placeholders, there's no need to explicitly pass values to variables when `s.run(...)`-ing\n",
    "* You can use variables the same way you use transformations \n",
    " "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Creating a shared variable\n",
    "shared_vector_1 = tf.Variable(initial_value=np.ones(5),\n",
    "                              name=\"example_variable\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial value [ 1.  1.  1.  1.  1.]\n"
     ]
    }
   ],
   "source": [
    "# Initialize variable(s) with initial values\n",
    "s.run(tf.global_variables_initializer())\n",
    "\n",
    "# Evaluating shared variable (outside symbolicd graph)\n",
    "print(\"Initial value\", s.run(shared_vector_1))\n",
    "\n",
    "# Within symbolic graph you use them just\n",
    "# as any other inout or transformation, not \"get value\" needed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "New value [ 0.  1.  2.  3.  4.]\n"
     ]
    }
   ],
   "source": [
    "# Setting a new value\n",
    "s.run(shared_vector_1.assign(np.arange(5)))\n",
    "\n",
    "# Getting that new value\n",
    "print(\"New value\", s.run(shared_vector_1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# tf.gradients - why graphs matter\n",
    "* Tensorflow can compute derivatives and gradients automatically using the computation graph\n",
    "* True to its name it can manage matrix derivatives\n",
    "* Gradients are computed as a product of elementary derivatives via the chain rule:\n",
    "\n",
    "$$ {\\partial f(g(x)) \\over \\partial x} = {\\partial f(g(x)) \\over \\partial g(x)}\\cdot {\\partial g(x) \\over \\partial x} $$\n",
    "\n",
    "It can get you the derivative of any graph as long as it knows how to differentiate elementary operations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "my_scalar = tf.placeholder('float32')\n",
    "\n",
    "scalar_squared = my_scalar**2\n",
    "\n",
    "# A derivative of scalar_squared by my_scalar\n",
    "derivative = tf.gradients(scalar_squared, [my_scalar, ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<tf.Tensor 'gradients/pow_1_grad/Reshape:0' shape=<unknown> dtype=float32>]"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "derivative"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAD8CAYAAABjAo9vAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9bfc761978>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "x = np.linspace(-3, 3)\n",
    "x_squared, x_squared_der = s.run([scalar_squared, derivative[0]],\n",
    "                                 {my_scalar:x})\n",
    "\n",
    "plt.plot(x, x_squared,label=\"$x^2$\")\n",
    "plt.plot(x, x_squared_der, label=r\"$\\frac{dx^2}{dx}$\")\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Why that rocks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "my_vector = tf.placeholder('float32', [None])\n",
    "# Compute the gradient of the next weird function over my_scalar and my_vector\n",
    "# Warning! Trying to understand the meaning of that function may result in permanent brain damage\n",
    "weird_psychotic_function = tf.reduce_mean(\n",
    "    (my_vector+my_scalar)**(1+tf.nn.moments(my_vector,[0])[1]) + \n",
    "    1./ tf.atan(my_scalar))/(my_scalar**2 + 1) + 0.01*tf.sin(\n",
    "    2*my_scalar**1.5)*(tf.reduce_sum(my_vector)* my_scalar**2\n",
    "                      )*tf.exp((my_scalar-4)**2)/(\n",
    "    1+tf.exp((my_scalar-4)**2))*(1.-(tf.exp(-(my_scalar-4)**2)\n",
    "                                    )/(1+tf.exp(-(my_scalar-4)**2)))**2\n",
    "\n",
    "der_by_scalar = tf.gradients(weird_psychotic_function, my_scalar)\n",
    "der_by_vector = tf.gradients(weird_psychotic_function, my_vector)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAD8CAYAAAB0IB+mAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJiTgCyFENyEBXwghuon/B/a5zgq1LoouAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9bf401edd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plotting the derivative\n",
    "scalar_space = np.linspace(1, 7, 100)\n",
    "\n",
    "y = [s.run(weird_psychotic_function, {my_scalar:x, my_vector:[1, 2, 3]})\n",
    "     for x in scalar_space]\n",
    "\n",
    "plt.plot(scalar_space, y, label='function')\n",
    "\n",
    "y_der_by_scalar = [s.run(der_by_scalar,\n",
    "                         {my_scalar:x, my_vector:[1, 2, 3]})\n",
    "                   for x in scalar_space]\n",
    "\n",
    "plt.plot(scalar_space, y_der_by_scalar, label='derivative')\n",
    "plt.grid()\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Almost done - optimizers\n",
    "\n",
    "While you can perform gradient descent by hand with automatic grads from above, tensorflow also has some optimization methods implemented for you. Recall momentum & rmsprop?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y_guess = tf.Variable(np.zeros(2, dtype='float32'))\n",
    "y_true = tf.range(1, 3, dtype='float32')\n",
    "loss = tf.reduce_mean((y_guess - y_true + tf.random_normal([2]))**2) \n",
    "#loss = tf.reduce_mean((y_guess - y_true)**2) \n",
    "optimizer = tf.train.MomentumOptimizer(0.01, 0.5).minimize(\n",
    "    loss, var_list=y_guess)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAAD8CAYAAACfF6SlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nXeHfRRdddPEyxP8HmmeFIdgYacwAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9becba8a58>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import animation, rc\n",
    "import matplotlib_utils\n",
    "from IPython.display import HTML, display_html\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "y_true_value = s.run(y_true)\n",
    "level_x = np.arange(0, 2, 0.02)\n",
    "level_y = np.arange(0, 3, 0.02)\n",
    "X, Y = np.meshgrid(level_x, level_y)\n",
    "Z = (X - y_true_value[0])**2 + (Y - y_true_value[1])**2\n",
    "ax.set_xlim(-0.02, 2)\n",
    "ax.set_ylim(-0.02, 3)\n",
    "s.run(tf.global_variables_initializer())\n",
    "ax.scatter(*s.run(y_true), c='red')\n",
    "contour = ax.contour(X, Y, Z, 10)\n",
    "ax.clabel(contour, inline=1, fontsize=10)\n",
    "line, = ax.plot([], [], lw=2)\n",
    "\n",
    "def init():\n",
    "    line.set_data([], [])\n",
    "    return (line,)\n",
    "\n",
    "guesses = [s.run(y_guess)]\n",
    "\n",
    "def animate(i):\n",
    "    s.run(optimizer)\n",
    "    guesses.append(s.run(y_guess))\n",
    "    line.set_data(*zip(*guesses))\n",
    "    return (line,)\n",
    "\n",
    "anim = animation.FuncAnimation(fig, animate, init_func=init,\n",
    "                               frames=400, interval=20, blit=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#try:\n",
    "#    display_html(HTML(anim.to_html5_video()))\n",
    "# In case the build-in renderers are unaviable, fall back to\n",
    "# a custom one, that doesn't require external libraries\n",
    "#except RuntimeError:\n",
    "#    anim.save(None, writer=matplotlib_utils.SimpleMovieWriter(0.001))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Logistic regression\n",
    "Your assignment is to implement the logistic regression\n",
    "\n",
    "Plan:\n",
    "* Use a shared variable for weights\n",
    "* Use a matrix placeholder for `X`\n",
    " \n",
    "We shall train on a two-class MNIST dataset\n",
    "* please note that target `y` are `{0,1}` and not `{-1,1}` as in some formulae"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y [shape - (360,)]: [0 1 0 1 0 1 0 0 1 1]\n",
      "X [shape - (360, 64)]:\n"
     ]
    }
   ],
   "source": [
    "from sklearn.datasets import load_digits\n",
    "mnist = load_digits(2)\n",
    "\n",
    "X, y = mnist.data, mnist.target\n",
    "\n",
    "print(\"y [shape - %s]:\" % (str(y.shape)), y[:10])\n",
    "print(\"X [shape - %s]:\" % (str(X.shape)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X:\n",
      " [[  0.   0.   5.  13.   9.   1.   0.   0.   0.   0.]\n",
      " [  0.   0.   0.  12.  13.   5.   0.   0.   0.   0.]\n",
      " [  0.   0.   1.   9.  15.  11.   0.   0.   0.   0.]]\n",
      "y:\n",
      " [0 1 0 1 0 1 0 0 1 1]\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPgAAAD8CAYAAABaQGkdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n9l+8Q5/pEyhkXAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9becba8438>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('X:\\n',X[:3,:10])\n",
    "print('y:\\n',y[:10])\n",
    "plt.imshow(X[0].reshape([8,8]));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It's your turn now!\n",
    "Just a small reminder of the relevant math:\n",
    "\n",
    "$$\n",
    "P(y=1|X) = \\sigma(X \\cdot W + b)\n",
    "$$\n",
    "$$\n",
    "\\text{loss} = -\\log\\left(P\\left(y_\\text{predicted} = 1\\right)\\right)\\cdot y_\\text{true} - \\log\\left(1 - P\\left(y_\\text{predicted} = 1\\right)\\right)\\cdot\\left(1 - y_\\text{true}\\right)\n",
    "$$\n",
    "\n",
    "$\\sigma(x)$ is available via `tf.nn.sigmoid` and matrix multiplication via `tf.matmul`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(\n",
    "    X, y, random_state=42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "__Your code goes here.__ For the training and testing scaffolding to work, please stick to the names in comments."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Model parameters - weights and bias\n",
    "weights = tf.Variable(tf.random_normal(shape=(X.shape[1], 1),stddev=0.1,dtype=tf.float32))\n",
    "b = tf.Variable(0,dtype=tf.float32)\n",
    "\n",
    "# Add an op to initialize the variables.\n",
    "s.run(tf.global_variables_initializer())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Placeholders for the input data\n",
    "input_X = tf.placeholder(tf.float32)\n",
    "input_y = tf.placeholder(tf.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# The model code\n",
    "# Compute a vector of predictions, resulting shape should be [input_X.shape[0],]\n",
    "# This is 1D, if you have extra dimensions, you can  get rid of them with tf.squeeze .\n",
    "predicted_y = tf.squeeze(tf.sigmoid(tf.matmul(input_X, weights)+b))\n",
    "loss = -tf.reduce_mean( input_y*tf.log(predicted_y) + (1-input_y)*tf.log(1-predicted_y) )\n",
    "optimizer = tf.train.AdagradOptimizer(0.01).minimize(loss=loss)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A test to help with the debugging"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "validation_weights = 1e-3 * np.fromiter(map(lambda x:\n",
    "        s.run(weird_psychotic_function, {my_scalar:x, my_vector:[1, 0.1, 2]}),\n",
    "                                   0.15 * np.arange(1, X.shape[1] + 1)),\n",
    "                                   count=X.shape[1], dtype=np.float32)[:, np.newaxis]\n",
    "# Compute predictions for given weights and bias\n",
    "prediction_validation = s.run(\n",
    "    predicted_y, {\n",
    "    input_X: X,\n",
    "    weights: validation_weights,\n",
    "    b: 1e-1})\n",
    "\n",
    "# Load the reference values for the predictions\n",
    "validation_true_values = np.loadtxt(\"validation_predictons.txt\")\n",
    "\n",
    "assert prediction_validation.shape == (X.shape[0],),\\\n",
    "       \"Predictions must be a 1D array with length equal to the number \" \\\n",
    "       \"of examples in input_X\"\n",
    "assert np.allclose(validation_true_values, prediction_validation)\n",
    "loss_validation = s.run(\n",
    "        loss, {\n",
    "            input_X: X[:100],\n",
    "            input_y: y[-100:],\n",
    "            weights: validation_weights+1.21e-3,\n",
    "            b: -1e-1})\n",
    "assert np.allclose(loss_validation, 0.728689)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loss at iter 0:0.5766\n",
      "train auc: 0.800395256917\n",
      "test auc: 0.880928853755\n",
      "loss at iter 1:0.3538\n",
      "train auc: 0.930994729908\n",
      "test auc: 0.960968379447\n",
      "loss at iter 2:0.2573\n",
      "train auc: 0.969422485727\n",
      "test auc: 0.978754940711\n",
      "loss at iter 3:0.2028\n",
      "train auc: 0.982927097058\n",
      "test auc: 0.991106719368\n",
      "loss at iter 4:0.1679\n",
      "train auc: 0.990173473869\n",
      "test auc: 0.994071146245\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import roc_auc_score\n",
    "s.run(tf.global_variables_initializer())\n",
    "for i in range(5):\n",
    "    s.run(optimizer, {input_X: X_train, input_y: y_train})\n",
    "    loss_i = s.run(loss, {input_X: X_train, input_y: y_train})\n",
    "    print(\"loss at iter %i:%.4f\" % (i, loss_i))\n",
    "    print(\"train auc:\", roc_auc_score(y_train, s.run(predicted_y, {input_X:X_train})))\n",
    "    print(\"test auc:\", roc_auc_score(y_test, s.run(predicted_y, {input_X:X_test})))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Coursera submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "grade_submitter = grading.Grader(\"BJCiiY8sEeeCnhKCj4fcOA\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "test_weights = 1e-3 * np.fromiter(map(lambda x:\n",
    "    s.run(weird_psychotic_function, {my_scalar:x, my_vector:[1, 2, 3]}),\n",
    "                               0.1 * np.arange(1, X.shape[1] + 1)),\n",
    "                               count=X.shape[1], dtype=np.float32)[:, np.newaxis]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, test prediction and loss computation. This part doesn't require a fitted model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "prediction_test = s.run(\n",
    "    predicted_y, {\n",
    "    input_X: X,\n",
    "    weights: test_weights,\n",
    "    b: 1e-1})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "assert prediction_test.shape == (X.shape[0],),\\\n",
    "       \"Predictions must be a 1D array with length equal to the number \" \\\n",
    "       \"of examples in X_test\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "grade_submitter.set_answer(\"0ENlN\", prediction_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "loss_test = s.run(\n",
    "    loss, {\n",
    "        input_X: X[:100],\n",
    "        input_y: y[-100:],\n",
    "        weights: test_weights+1.21e-3,\n",
    "        b: -1e-1})\n",
    "# Yes, the X/y indices mistmach is intentional"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "grade_submitter.set_answer(\"mMVpM\", loss_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "grade_submitter.set_answer(\"D16Rc\", roc_auc_score(y_test, s.run(predicted_y, {input_X:X_test})))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Please use the credentials obtained from the Coursera assignment page."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Submitted to Coursera platform. See results on assignment page!\n"
     ]
    }
   ],
   "source": [
    "grade_submitter.submit(email, 'eTjLsFsOgFoZN4rg')"
   ]
  },
  {
   "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": "3.6.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}