A jupyter notebook explaining the basic usage of Tensorflow

TensorFlow Basics

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from __future__ import absolute_import, division, print_function\n",
    "\n",
    "import tensorflow as tf\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tensorflow Basics\n",
    "This tutorial should give a brief overview over the basic functionality of Tensorflow (TF).\n",
    "In TF computations are represented as graphs. A graph consists of nodes, which represent computations with Tensors. The graph is evaluated in a Session."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Graphs\n",
    "Generally the first step in writing a TF program is constructing the computation graph. When you import TF, a default graph is created. It is also possible to manually create different indepent graphs. Let's add some operations to the graph:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"Add_24:0\", shape=(), dtype=int32)\n",
      "Tensor(\"Add_25:0\", shape=(9,), dtype=int64)\n",
      "Tensor(\"MatMul:0\", shape=(3, 3), dtype=float32)\n",
      "Tensor(\"Reshape:0\", shape=(9,), dtype=float32)\n"
     ]
    }
   ],
   "source": [
    "# operations added to default graph\n",
    "add_int = tf.add(1,3)\n",
    "\n",
    "array = np.arange(9)\n",
    "add_array = tf.add(array,array)\n",
    "\n",
    "# define new graph\n",
    "new_graph = tf.Graph()\n",
    "with new_graph.as_default():\n",
    "    tensor = tf.constant([i for i in range(1,10)], dtype=tf.float32, shape=[3,3])\n",
    "    mult_tensor =tf.matmul(tensor,tensor)\n",
    "\n",
    "    reshape_tensor = tf.reshape(tensor, shape=[-1])\n",
    "\n",
    "print(add_int)\n",
    "print(add_array) \n",
    "print(mult_tensor) \n",
    "print(reshape_tensor)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Usually operations can take any tensor argument, that is accepted by [tf.convert_to_tensor](https://www.tensorflow.org/versions/r0.11/api_docs/python/framework.html#convert_to_tensor). There are a lot of defined operation you can use on tensors like [math](https://www.tensorflow.org/versions/r0.11/api_docs/python/math_ops.html#math), [tensor transformation](https://www.tensorflow.org/versions/r0.11/api_docs/python/array_ops.html#tensor-transformations), [special neural network](https://www.tensorflow.org/versions/r0.11/api_docs/python/nn.html#neural-network) operations and many more."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In order to evaluate the operations, we have to launch the graph in a session:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating operations on default graph:\n",
      "4\n",
      "[ 0  2  4  6  8 10 12 14 16]\n",
      "<type 'numpy.ndarray'>\n",
      "\n",
      "Trying to evaluate operation on another graph:\n",
      "Fetch argument <tf.Tensor 'MatMul:0' shape=(3, 3) dtype=float32> cannot be interpreted as a Tensor. (Tensor Tensor(\"MatMul:0\", shape=(3, 3), dtype=float32) is not an element of this graph.)\n",
      "\n",
      "Now the operations on the other graph can be evaluated\n",
      "[[  30.   36.   42.]\n",
      " [  66.   81.   96.]\n",
      " [ 102.  126.  150.]]\n",
      "[ 1.  2.  3.  4.  5.  6.  7.  8.  9.]\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    print('Evaluating operations on default graph:')\n",
    "    print(sess.run(add_int))\n",
    "    print(sess.run(add_array))\n",
    "    print(type(sess.run(add_array)))\n",
    "    print('\\nTrying to evaluate operation on another graph:')\n",
    "    try:\n",
    "        print(sess.run(mult_tensor))\n",
    "    except Exception as e:\n",
    "        print(e)\n",
    "        \n",
    "with tf.Session(graph=new_graph)as sess:\n",
    "    print('\\nNow the operations on the other graph can be evaluated')\n",
    "    print(sess.run(mult_tensor))\n",
    "    print(sess.run(reshape_tensor))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Beside tf.constant, there are other common data structures to represent data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Var:\n",
      "[[ 0.30460876  1.07989275]\n",
      " [ 0.12586282 -1.57130408]]\n",
      "\n",
      "Operation:\n",
      "[[ 1.30460882  2.07989264]\n",
      " [ 1.12586284 -0.57130408]]\n",
      "\n",
      " Cannot feed value of shape (3, 3) for Tensor u'Placeholder_13:0', which has shape '(2, 2)'\n"
     ]
    }
   ],
   "source": [
    "var = tf.Variable(tf.random_normal(shape=[2,2]))\n",
    "plc = tf.placeholder(dtype=tf.float32, shape=[2,2])\n",
    "operation = tf.add(var,plc)\n",
    "\n",
    "feed_data = np.ones((2,2))\n",
    "feed_data2 = np.ones((3,3))\n",
    "\n",
    "init = tf.initialize_all_variables()\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    sess.run(init)\n",
    "    print('Var:')\n",
    "    print(sess.run(var))\n",
    "    print('\\nOperation:')\n",
    "    print(sess.run(operation, feed_dict={plc: feed_data}))\n",
    "    try:\n",
    "        print(sess.run(operation, feed_dict={plc: feed_data2}))\n",
    "    except Exception as e:\n",
    "        print('\\n', e)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A [tf.Variable](https://www.tensorflow.org/versions/r0.11/api_docs/python/state_ops.html#Variable) has to initialized via the init operation. Variables are used to store parameters of a model. They can also be saved in a file.\n",
    "The [tf.placeholder](https://www.tensorflow.org/versions/r0.11/api_docs/python/io_ops.html#placeholder) is used to feed data inside the model. They can initialized to have a certain shape, to make sure only suitable data is fed into the computation."
   ]
  }
 ],
 "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": 1
}