|
2 | 2 | "cells": [ |
3 | 3 | { |
4 | 4 | "cell_type": "code", |
5 | | - "execution_count": 187, |
| 5 | + "execution_count": 193, |
6 | 6 | "metadata": { |
7 | 7 | "collapsed": false |
8 | 8 | }, |
9 | 9 | "outputs": [], |
10 | 10 | "source": [ |
11 | 11 | "import numpy as np\n", |
12 | 12 | "import random\n", |
13 | | - "from collections import Counter\n", |
14 | | - "\n", |
15 | 13 | "\n", |
16 | 14 | "def create_feature_sets_and_labels(test_size = 0.2):\n", |
17 | 15 | "\n", |
18 | | - " # 5 features\n", |
| 16 | + " # known patterns (5 features) output of [1] of positions [0,4]==1\n", |
19 | 17 | " features = []\n", |
20 | 18 | " features.append([[0, 0, 0, 0, 0], [0,1]])\n", |
21 | 19 | " features.append([[0, 0, 0, 0, 1], [0,1]])\n", |
|
25 | 23 | " features.append([[1, 1, 1, 1, 0], [0,1]])\n", |
26 | 24 | " features.append([[1, 1, 1, 0, 0], [0,1]])\n", |
27 | 25 | " features.append([[1, 1, 0, 0, 0], [0,1]])\n", |
28 | | - "# features.append([[1, 0, 0, 0, 0], [0,1]])\n", |
29 | | - "# features.append([[1, 0, 0, 1, 0], [0,1]])\n", |
30 | | - "# features.append([[1, 0, 1, 1, 0], [0,1]])\n", |
31 | | - "# features.append([[1, 1, 0, 1, 0], [0,1]])\n", |
32 | | - "# features.append([[0, 1, 0, 1, 1], [0,1]])\n", |
33 | | - "# features.append([[0, 0, 1, 0, 1], [0,1]])\n", |
34 | 26 | "\n", |
35 | | - " # output of [1] of positions [0,4]==1\n", |
36 | 27 | " features.append([[1, 0, 0, 0, 1], [1,0]])\n", |
37 | 28 | " features.append([[1, 1, 0, 0, 1], [1,0]])\n", |
38 | 29 | " features.append([[1, 1, 1, 0, 1], [1,0]])\n", |
39 | 30 | " features.append([[1, 1, 1, 1, 1], [1,0]])\n", |
40 | 31 | " features.append([[1, 0, 0, 1, 1], [1,0]])\n", |
41 | | - "# features.append([[1, 0, 1, 1, 1], [1,0]])\n", |
42 | | - "# features.append([[1, 1, 0, 1, 1], [1,0]])\n", |
43 | | - "# features.append([[1, 0, 1, 0, 1], [1,0]])\n", |
44 | 32 | "\n", |
| 33 | + "# unknown patterns\n", |
| 34 | + "# features.append([[1, 0, 0, 0, 0], [0,1]])\n", |
| 35 | + "# features.append([[1, 0, 0, 1, 0], [0,1]])\n", |
| 36 | + "# features.append([[1, 0, 1, 1, 0], [0,1]])\n", |
| 37 | + "# features.append([[1, 1, 0, 1, 0], [0,1]])\n", |
| 38 | + "# features.append([[0, 1, 0, 1, 1], [0,1]])\n", |
| 39 | + "# features.append([[0, 0, 1, 0, 1], [0,1]])\n", |
| 40 | + "# features.append([[1, 0, 1, 1, 1], [1,0]])\n", |
| 41 | + "# features.append([[1, 1, 0, 1, 1], [1,0]])\n", |
| 42 | + "# features.append([[1, 0, 1, 0, 1], [1,0]])\n", |
| 43 | + "\n", |
| 44 | + " # shuffle out features and turn into np.array\n", |
45 | 45 | " random.shuffle(features)\n", |
46 | 46 | " features = np.array(features)\n", |
47 | 47 | "\n", |
| 48 | + " # split a portion of the features into tests\n", |
48 | 49 | " testing_size = int(test_size*len(features))\n", |
49 | 50 | "\n", |
| 51 | + " # create train and test lists\n", |
50 | 52 | " train_x = list(features[:,0][:-testing_size])\n", |
51 | 53 | " train_y = list(features[:,1][:-testing_size])\n", |
52 | 54 | " test_x = list(features[:,0][-testing_size:])\n", |
53 | 55 | " test_y = list(features[:,1][-testing_size:])\n", |
54 | 56 | "\n", |
55 | | - " return train_x,train_y,test_x,test_y\n", |
56 | | - "\n", |
57 | | - "if __name__ == '__main__':\n", |
58 | | - " train_x,train_y,test_x,test_y = create_feature_sets_and_labels()\n" |
| 57 | + " return train_x,train_y,test_x,test_y\n" |
59 | 58 | ] |
60 | 59 | }, |
61 | 60 | { |
|
64 | 63 | "metadata": { |
65 | 64 | "collapsed": false |
66 | 65 | }, |
67 | | - "outputs": [ |
68 | | - { |
69 | | - "name": "stdout", |
70 | | - "output_type": "stream", |
71 | | - "text": [ |
72 | | - "Epoch 10 completed out of 50 cost: 0.3681\n", |
73 | | - "Epoch 20 completed out of 50 cost: 0.0166706\n", |
74 | | - "Epoch 30 completed out of 50 cost: 0.00777522\n", |
75 | | - "Epoch 40 completed out of 50 cost: 0.00493255\n", |
76 | | - "Accuracy: 1.0\n", |
77 | | - "[[-0.26135445 4.12279558]]\n" |
78 | | - ] |
79 | | - } |
80 | | - ], |
| 66 | + "outputs": [], |
81 | 67 | "source": [ |
82 | 68 | "import tensorflow as tf\n", |
83 | 69 | "import numpy as np\n", |
84 | 70 | "\n", |
85 | 71 | "train_x,train_y,test_x,test_y = create_feature_sets_and_labels()\n", |
86 | 72 | "\n", |
| 73 | + "# hidden layers and their nodes\n", |
87 | 74 | "n_nodes_hl1 = 20\n", |
88 | 75 | "n_nodes_hl2 = 20\n", |
89 | 76 | "\n", |
| 77 | + "# classes in our output\n", |
90 | 78 | "n_classes = 2\n", |
| 79 | + "# iterations and batch-size to build out model\n", |
91 | 80 | "hm_epochs = 50\n", |
92 | 81 | "batch_size = 4\n", |
93 | 82 | "\n", |
94 | 83 | "x = tf.placeholder('float')\n", |
95 | 84 | "y = tf.placeholder('float')\n", |
96 | 85 | "\n", |
| 86 | + "# random weights and bias for our layers\n", |
97 | 87 | "hidden_1_layer = {'f_fum':n_nodes_hl1,\n", |
98 | 88 | " 'weight':tf.Variable(tf.random_normal([len(train_x[0]), n_nodes_hl1])),\n", |
99 | 89 | " 'bias':tf.Variable(tf.random_normal([n_nodes_hl1]))}\n", |
|
107 | 97 | " 'bias':tf.Variable(tf.random_normal([n_classes])),}\n", |
108 | 98 | " \n", |
109 | 99 | "\n", |
| 100 | + "# our predictive model's definition\n", |
110 | 101 | "def neural_network_model(data):\n", |
111 | 102 | "\n", |
112 | 103 | " l1 = tf.add(tf.matmul(data,hidden_1_layer['weight']), hidden_1_layer['bias'])\n", |
|
119 | 110 | "\n", |
120 | 111 | " return output\n", |
121 | 112 | "\n", |
| 113 | + "# training our model\n", |
122 | 114 | "def train_neural_network(x):\n", |
| 115 | + " # use the model definition\n", |
123 | 116 | " prediction = neural_network_model(x)\n", |
124 | 117 | "\n", |
| 118 | + " # formula for cost (error)\n", |
125 | 119 | " cost = tf.reduce_mean( tf.nn.softmax_cross_entropy_with_logits(prediction,y) )\n", |
126 | | - " #optimizer = tf.train.AdamOptimizer(learning_rate=0.0001, beta1=0.9, beta2=0.999).minimize(cost)\n", |
| 120 | + " # optimize for cost using GradientDescent\n", |
127 | 121 | " optimizer = tf.train.GradientDescentOptimizer(1).minimize(cost)\n", |
128 | 122 | "\n", |
| 123 | + " # Tensorflow session\n", |
129 | 124 | " with tf.Session() as sess:\n", |
| 125 | + " # initialize our variables\n", |
130 | 126 | " sess.run(tf.global_variables_initializer())\n", |
131 | | - " \n", |
| 127 | + "\n", |
| 128 | + " # loop through specified number of iterations\n", |
132 | 129 | " for epoch in range(hm_epochs):\n", |
133 | 130 | " epoch_loss = 0\n", |
134 | 131 | " i=0\n", |
| 132 | + " # handle batch sized chunks of training data\n", |
135 | 133 | " while i < len(train_x):\n", |
136 | 134 | " start = i\n", |
137 | 135 | " end = i+batch_size\n", |
|
143 | 141 | " i+=batch_size\n", |
144 | 142 | " last_cost = c\n", |
145 | 143 | "\n", |
146 | | - " if (epoch% 10) == 0 and epoch > 1:\n", |
| 144 | + " # print cost updates along the way\n", |
| 145 | + " if (epoch% (hm_epochs/5)) == 0:\n", |
147 | 146 | " print('Epoch', epoch, 'completed out of',hm_epochs,'cost:', last_cost)\n", |
148 | 147 | "\n", |
| 148 | + " # print accuracy of our model against known test data (a subset of the shuffled known patterns)\n", |
149 | 149 | " correct = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))\n", |
150 | 150 | " accuracy = tf.reduce_mean(tf.cast(correct, 'float'))\n", |
151 | 151 | " print('Accuracy:',accuracy.eval({x:test_x, y:test_y}))\n", |
152 | 152 | "\n", |
| 153 | + " # print a prediction using our model\n", |
153 | 154 | " output = prediction.eval(feed_dict = {x: [[1, 0, 0, 0, 0]]})\n", |
154 | 155 | " print(output)\n", |
| 156 | + " # normalize the prediction values\n", |
155 | 157 | " print(tf.sigmoid(output[0][0]).eval(), tf.sigmoid(output[0][1]).eval())\n", |
156 | 158 | " \n", |
157 | 159 | "train_neural_network(x)\n" |
|
0 commit comments