Skip to content

Instantly share code, notes, and snippets.

@analyticsindiamagazine

analyticsindiamagazine/Intel i5-4210U MNIST.ipynb

Last active June 26, 2018 06:37
Show Gist options
  • Save analyticsindiamagazine/4e3fe007ff0f3ba2257ac21958996b88 to your computer and use it in GitHub Desktop.
Save analyticsindiamagazine/4e3fe007ff0f3ba2257ac21958996b88 to your computer and use it in GitHub Desktop.
Download ZIP
Display the source blob
Display the rendered blob
Raw
Intel i5-4210U MNIST.ipynb
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Tensorflow MNIST- Intel i5-4210U"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[name: \"/device:CPU:0\"\n",
"device_type: \"CPU\"\n",
"memory_limit: 268435456\n",
"locality {\n",
"}\n",
"incarnation: 2983841888649849212\n",
"]\n"
]
}
],
"source": [
"import tensorflow\n",
"from tensorflow.python.client import device_lib\n",
"print(device_lib.list_local_devices())"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"WARNING:tensorflow:From <ipython-input-3-acd6a8829fd9>:14: read_data_sets (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
"WARNING:tensorflow:From c:\\users\\kishan maladkar\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:260: maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Please write your own downloading logic.\n",
"WARNING:tensorflow:From c:\\users\\kishan maladkar\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:262: extract_images (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Please use tf.data to implement this functionality.\n",
"Extracting data/MNIST/train-images-idx3-ubyte.gz\n",
"WARNING:tensorflow:From c:\\users\\kishan maladkar\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:267: extract_labels (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Please use tf.data to implement this functionality.\n",
"Extracting data/MNIST/train-labels-idx1-ubyte.gz\n",
"WARNING:tensorflow:From c:\\users\\kishan maladkar\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:110: dense_to_one_hot (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Please use tf.one_hot on tensors.\n",
"Extracting data/MNIST/t10k-images-idx3-ubyte.gz\n",
"Extracting data/MNIST/t10k-labels-idx1-ubyte.gz\n",
"WARNING:tensorflow:From c:\\users\\kishan maladkar\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:290: DataSet.__init__ (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
"WARNING:tensorflow:From <ipython-input-3-acd6a8829fd9>:148: softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"\n",
"Future major versions of TensorFlow will allow gradients to flow\n",
"into the labels input on backprop by default.\n",
"\n",
"See @{tf.nn.softmax_cross_entropy_with_logits_v2}.\n",
"\n",
"Optimization Iteration: 1, Training Accuracy: 4.7%\n",
"Optimization Iteration: 101, Training Accuracy: 81.2%\n",
"Optimization Iteration: 201, Training Accuracy: 87.5%\n",
"Optimization Iteration: 301, Training Accuracy: 85.9%\n",
"Optimization Iteration: 401, Training Accuracy: 95.3%\n",
"Optimization Iteration: 501, Training Accuracy: 92.2%\n",
"Optimization Iteration: 601, Training Accuracy: 90.6%\n",
"Optimization Iteration: 701, Training Accuracy: 95.3%\n",
"Optimization Iteration: 801, Training Accuracy: 90.6%\n",
"Optimization Iteration: 901, Training Accuracy: 89.1%\n",
"Optimization Iteration: 1001, Training Accuracy: 95.3%\n",
"Optimization Iteration: 1101, Training Accuracy: 95.3%\n",
"Optimization Iteration: 1201, Training Accuracy: 95.3%\n",
"Optimization Iteration: 1301, Training Accuracy: 100.0%\n",
"Optimization Iteration: 1401, Training Accuracy: 96.9%\n",
"Optimization Iteration: 1501, Training Accuracy: 98.4%\n",
"Optimization Iteration: 1601, Training Accuracy: 96.9%\n",
"Optimization Iteration: 1701, Training Accuracy: 96.9%\n",
"Optimization Iteration: 1801, Training Accuracy: 95.3%\n",
"Optimization Iteration: 1901, Training Accuracy: 98.4%\n",
"Optimization Iteration: 2001, Training Accuracy: 95.3%\n",
"Optimization Iteration: 2101, Training Accuracy: 100.0%\n",
"Optimization Iteration: 2201, Training Accuracy: 96.9%\n",
"Optimization Iteration: 2301, Training Accuracy: 96.9%\n",
"Optimization Iteration: 2401, Training Accuracy: 96.9%\n",
"Optimization Iteration: 2501, Training Accuracy: 98.4%\n",
"Optimization Iteration: 2601, Training Accuracy: 95.3%\n",
"Optimization Iteration: 2701, Training Accuracy: 96.9%\n",
"Optimization Iteration: 2801, Training Accuracy: 98.4%\n",
"Optimization Iteration: 2901, Training Accuracy: 98.4%\n",
"Optimization Iteration: 3001, Training Accuracy: 96.9%\n",
"Optimization Iteration: 3101, Training Accuracy: 93.8%\n",
"Optimization Iteration: 3201, Training Accuracy: 100.0%\n",
"Optimization Iteration: 3301, Training Accuracy: 100.0%\n",
"Optimization Iteration: 3401, Training Accuracy: 98.4%\n",
"Optimization Iteration: 3501, Training Accuracy: 100.0%\n",
"Optimization Iteration: 3601, Training Accuracy: 95.3%\n",
"Optimization Iteration: 3701, Training Accuracy: 98.4%\n",
"Optimization Iteration: 3801, Training Accuracy: 96.9%\n",
"Optimization Iteration: 3901, Training Accuracy: 98.4%\n",
"Optimization Iteration: 4001, Training Accuracy: 98.4%\n",
"Optimization Iteration: 4101, Training Accuracy: 98.4%\n",
"Optimization Iteration: 4201, Training Accuracy: 96.9%\n",
"Optimization Iteration: 4301, Training Accuracy: 100.0%\n",
"Optimization Iteration: 4401, Training Accuracy: 96.9%\n",
"Optimization Iteration: 4501, Training Accuracy: 100.0%\n",
"Optimization Iteration: 4601, Training Accuracy: 96.9%\n",
"Optimization Iteration: 4701, Training Accuracy: 100.0%\n",
"Optimization Iteration: 4801, Training Accuracy: 100.0%\n",
"Optimization Iteration: 4901, Training Accuracy: 95.3%\n",
"Optimization Iteration: 5001, Training Accuracy: 98.4%\n",
"Optimization Iteration: 5101, Training Accuracy: 92.2%\n",
"Optimization Iteration: 5201, Training Accuracy: 100.0%\n",
"Optimization Iteration: 5301, Training Accuracy: 98.4%\n",
"Optimization Iteration: 5401, Training Accuracy: 100.0%\n",
"Optimization Iteration: 5501, Training Accuracy: 100.0%\n",
"Optimization Iteration: 5601, Training Accuracy: 100.0%\n",
"Optimization Iteration: 5701, Training Accuracy: 98.4%\n",
"Optimization Iteration: 5801, Training Accuracy: 100.0%\n",
"Optimization Iteration: 5901, Training Accuracy: 96.9%\n",
"Optimization Iteration: 6001, Training Accuracy: 98.4%\n",
"Optimization Iteration: 6101, Training Accuracy: 96.9%\n",
"Optimization Iteration: 6201, Training Accuracy: 100.0%\n",
"Optimization Iteration: 6301, Training Accuracy: 98.4%\n",
"Optimization Iteration: 6401, Training Accuracy: 100.0%\n",
"Optimization Iteration: 6501, Training Accuracy: 100.0%\n",
"Optimization Iteration: 6601, Training Accuracy: 100.0%\n",
"Optimization Iteration: 6701, Training Accuracy: 93.8%\n",
"Optimization Iteration: 6801, Training Accuracy: 100.0%\n",
"Optimization Iteration: 6901, Training Accuracy: 100.0%\n",
"Optimization Iteration: 7001, Training Accuracy: 96.9%\n",
"Optimization Iteration: 7101, Training Accuracy: 100.0%\n",
"Optimization Iteration: 7201, Training Accuracy: 100.0%\n",
"Optimization Iteration: 7301, Training Accuracy: 98.4%\n",
"Optimization Iteration: 7401, Training Accuracy: 98.4%\n",
"Optimization Iteration: 7501, Training Accuracy: 100.0%\n",
"Optimization Iteration: 7601, Training Accuracy: 98.4%\n",
"Optimization Iteration: 7701, Training Accuracy: 98.4%\n",
"Optimization Iteration: 7801, Training Accuracy: 100.0%\n",
"Optimization Iteration: 7901, Training Accuracy: 100.0%\n",
"Optimization Iteration: 8001, Training Accuracy: 100.0%\n",
"Optimization Iteration: 8101, Training Accuracy: 98.4%\n",
"Optimization Iteration: 8201, Training Accuracy: 98.4%\n",
"Optimization Iteration: 8301, Training Accuracy: 96.9%\n",
"Optimization Iteration: 8401, Training Accuracy: 98.4%\n",
"Optimization Iteration: 8501, Training Accuracy: 100.0%\n",
"Optimization Iteration: 8601, Training Accuracy: 100.0%\n",
"Optimization Iteration: 8701, Training Accuracy: 98.4%\n",
"Optimization Iteration: 8801, Training Accuracy: 100.0%\n",
"Optimization Iteration: 8901, Training Accuracy: 100.0%\n",
"Optimization Iteration: 9001, Training Accuracy: 95.3%\n",
"Optimization Iteration: 9101, Training Accuracy: 98.4%\n",
"Optimization Iteration: 9201, Training Accuracy: 100.0%\n",
"Optimization Iteration: 9301, Training Accuracy: 100.0%\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Optimization Iteration: 9401, Training Accuracy: 100.0%\n",
"Optimization Iteration: 9501, Training Accuracy: 98.4%\n",
"Optimization Iteration: 9601, Training Accuracy: 98.4%\n",
"Optimization Iteration: 9701, Training Accuracy: 100.0%\n",
"Optimization Iteration: 9801, Training Accuracy: 100.0%\n",
"Optimization Iteration: 9901, Training Accuracy: 100.0%\n",
"Time usage: 0:27:52\n"
]
}
],
"source": [
"import os\n",
"import sys\n",
"os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"1\"\n",
"import numpy as np\n",
"import matplotlib.pyplot as plt \n",
"import tensorflow as tf\n",
"from sklearn.metrics import confusion_matrix\n",
"import time \n",
"from datetime import timedelta\n",
"import math\n",
"import pandas as pd\n",
"\n",
"from tensorflow.examples.tutorials.mnist import input_data\n",
"data = input_data.read_data_sets('data/MNIST/', one_hot=True)\n",
"\n",
"weight_matrix_size1 = 5 #5x5 pixcels\n",
"depth1 = 16 #16 depth\n",
"\n",
"weight_matrix_size2 = 5 #5x5 pixcels\n",
"depth2 = 32 #32 depth\n",
"\n",
"fully_conn_layer = 256 #neuros at end of fully connected layer\n",
"\n",
"#Data dimensions\n",
"\n",
"#We have an input image of 28 x 28 dimensions\n",
"img_size = 28\n",
"\n",
"# We have a one hot encoded matrix of length 28*28 = 784\n",
"img_size_flat = img_size * img_size\n",
"\n",
"#Shape of the image represented by\n",
"img_shape = (img_size,img_size)\n",
"\n",
"#Number of channels in the input image\n",
"num_channels = 1\n",
"\n",
"#Number of output classes to be trained on\n",
"num_classes = 10\n",
"\n",
"def weight_matrix(dimensions):\n",
" return tf.Variable(tf.truncated_normal(shape = dimensions, stddev=0.1))\n",
"def biases_matrix(length):\n",
" return tf.Variable(tf.constant(0.1,shape=[length]))\n",
"\n",
"#Helper functions for ConvNet\n",
"\n",
"def convolutional_layer(input, #The images\n",
" depth, #channels of the image\n",
" no_filters, #number of filters in the output\n",
" weight_matrix_size):\n",
" \n",
" dimensions = [weight_matrix_size,weight_matrix_size, depth, no_filters]\n",
" \n",
" weights = weight_matrix(dimensions)\n",
" \n",
" biases = biases_matrix(length=no_filters)\n",
" \n",
" layer = tf.nn.conv2d(input=input,\n",
" filter= weights,\n",
" strides=[1, 1, 1, 1], #stride 2\n",
" padding='SAME') #input size = output size\n",
" layer += biases\n",
" \n",
" layer = tf.nn.max_pool(value=layer,\n",
" ksize=[1, 2, 2, 1],\n",
" strides=[1, 2, 2, 1],\n",
" padding='SAME')\n",
" #Passing the pooled layer into ReLU Activation function\n",
" layer = tf.nn.relu(layer)\n",
" \n",
" return layer , weights\n",
"\n",
"# Helper function for Flattening the layer\n",
"\n",
"def flatten_layer(layer):\n",
" \n",
" layer_shape = layer.get_shape()\n",
" \n",
" num_features = layer_shape[1:4].num_elements()\n",
" \n",
" layer_flat = tf.reshape(layer,[-1,num_features])\n",
" \n",
" return layer_flat, num_features\n",
"\n",
"#Helper functions for activation and fully connected\n",
"\n",
"def fully_connected(input,num_inputs,\n",
" num_outputs,\n",
" use_relu = True):\n",
" weights = weight_matrix([num_inputs,num_outputs])\n",
" \n",
" biases = biases_matrix(length= num_outputs)\n",
" \n",
" layer = tf.matmul(input,weights) + biases\n",
" \n",
" if use_relu:\n",
" layer = tf.nn.relu(layer)\n",
" \n",
" return layer\n",
"\n",
"#Placeholder variables\n",
"\n",
"x = tf.placeholder(tf.float32,shape=[None,img_size_flat],name='x')\n",
"\n",
"x_image = tf.reshape(x, [-1,img_size,img_size,num_channels])\n",
"\n",
"y_true = tf.placeholder(tf.float32, shape=[None, num_classes], name = 'y_true')\n",
"\n",
"y_true_cls = tf.argmax(y_true, axis=1)\n",
"\n",
"# Setting up the network\n",
"\n",
"layer_conv1 , weights_conv1 = convolutional_layer(input = x_image,\n",
" depth = num_channels,\n",
" weight_matrix_size = weight_matrix_size1,\n",
" no_filters = depth1)\n",
"\n",
"#layer_conv1 shape = (-1,14,14,16) and dtype = float32\n",
"\n",
"layer_conv2 , weights_conv2 = convolutional_layer(input = layer_conv1,\n",
" depth = depth1,\n",
" weight_matrix_size = weight_matrix_size2,\n",
" no_filters = depth2)\n",
"#layer_conv2 = shape=(?, 7, 7, 36) dtype=float32\n",
"\n",
"#Flattening the layer\n",
"\n",
"layer_flat , num_features = flatten_layer(layer_conv2)\n",
"\n",
"#Fully connected layers\n",
"\n",
"layer_fc1 = fully_connected(input = layer_flat,\n",
" num_inputs = num_features,\n",
" num_outputs = fully_conn_layer,\n",
" use_relu = True)\n",
"\n",
"layer_fc2 = fully_connected(input = layer_fc1,\n",
" num_inputs = fully_conn_layer,\n",
" num_outputs = num_classes,\n",
" use_relu = False)\n",
"\n",
"y_pred = tf.nn.softmax(layer_fc2)\n",
"\n",
"y_pred_cls = tf.argmax(y_pred , axis =1)\n",
"\n",
"cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=layer_fc2,\n",
" labels=y_true)\n",
"cost = tf.reduce_mean(cross_entropy)\n",
"\n",
"#optimizing cost function\n",
"\n",
"optimizer = tf.train.AdamOptimizer(learning_rate= 1e-4).minimize(cost)\n",
"\n",
"correct_prediction = tf.equal(y_pred_cls, y_true_cls)\n",
"\n",
"accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))\n",
"\n",
"#TensorFlow session \n",
"config = tf.ConfigProto(\n",
" device_count = {'gpu': 0}\n",
" )\n",
"session = tf.Session(config=config)\n",
"\n",
"session.run(tf.global_variables_initializer())\n",
"\n",
"train_batch_size = 64\n",
"\n",
"total_iterations = 0\n",
"\n",
"accuracy_ = tf.summary.scalar('accuracy_value', accuracy)\n",
"loss_ = tf.summary.scalar('loss_value', cost)\n",
"\n",
"def optimize(num_iterations):\n",
" \n",
" global total_iterations\n",
"\n",
" start_time = time.time()\n",
" \n",
" summary_op = tf.summary.merge_all()\n",
"\n",
" file_writer = tf.summary.FileWriter('/path/to/logs', session.graph)\n",
" for i in range(total_iterations,\n",
" total_iterations + num_iterations):\n",
" \n",
" x_batch, y_true_batch = data.train.next_batch(train_batch_size)\n",
" \n",
" feed_dict_train = {x: x_batch,\n",
" y_true: y_true_batch}\n",
" \n",
" session.run(optimizer, feed_dict=feed_dict_train)\n",
" \n",
" acc_value = session.run(accuracy_, feed_dict=feed_dict_train)\n",
" loss_value = session.run(loss_, feed_dict=feed_dict_train)\n",
" file_writer.add_summary(acc_value, i)\n",
" file_writer.add_summary(loss_value, i)\n",
" \n",
" \n",
" if i % 100 == 0:\n",
" \n",
" acc = session.run(accuracy, feed_dict=feed_dict_train)\n",
" \n",
" msg = \"Optimization Iteration: {0:>6}, Training Accuracy: {1:>6.1%}\"\n",
" \n",
" print(msg.format(i + 1, acc))\n",
" \n",
" total_iterations += num_iterations\n",
" \n",
" \n",
" # Ending time.\n",
" end_time = time.time()\n",
"\n",
" # Difference between start and end-times.\n",
" time_dif = end_time - start_time\n",
"\n",
" # Print the time-usage.\n",
" print(\"Time usage: \" + str(timedelta(seconds=int(round(time_dif)))))\n",
" \n",
"optimize(num_iterations=10000)"
]
}
],
"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": "3.6.3"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment