|
| 1 | +# Lab 9 XOR |
| 2 | +# This example does not work |
| 3 | +import tensorflow as tf |
| 4 | +import numpy as np |
| 5 | +tf.set_random_seed(777) # for reproducibility |
| 6 | + |
| 7 | +x_data = np.array([[0, 0], [0, 1], [1, 0], [1, 1]], dtype=np.float32) |
| 8 | +y_data = np.array([[0], [1], [1], [0]], dtype=np.float32) |
| 9 | + |
| 10 | +X = tf.placeholder(tf.float32, [None, 2], name='x-input') |
| 11 | +Y = tf.placeholder(tf.float32, [None, 1], name='y-input') |
| 12 | + |
| 13 | +W1 = tf.Variable(tf.random_normal([2, 2]), name='weight1') |
| 14 | +b1 = tf.Variable(tf.random_normal([2]), name='bias1') |
| 15 | + |
| 16 | +with tf.name_scope("layer2") as scope: |
| 17 | + layer1 = tf.sigmoid(tf.matmul(X, W1) + b1) |
| 18 | + |
| 19 | +W2 = tf.Variable(tf.random_normal([2, 1]), name='weight2') |
| 20 | +b2 = tf.Variable(tf.random_normal([1]), name='bias2') |
| 21 | + |
| 22 | +with tf.name_scope("layer3") as scope: |
| 23 | + hypothesis = tf.sigmoid(tf.matmul(layer1, W2) + b2) |
| 24 | + |
| 25 | +w1_hist = tf.summary.histogram("weights1", W1) |
| 26 | +w2_hist = tf.summary.histogram("weights2", W2) |
| 27 | +b1_hist = tf.summary.histogram("biases1", b1) |
| 28 | +b2_hist = tf.summary.histogram("biases2", b2) |
| 29 | +y_hist = tf.summary.histogram("y", Y) |
| 30 | + |
| 31 | +# cost/loss function |
| 32 | +with tf.name_scope("cost") as scope: |
| 33 | + cost = -tf.reduce_mean(Y * tf.log(hypothesis) + (1 - Y) * |
| 34 | + tf.log(1 - hypothesis)) |
| 35 | + cost_summ = tf.summary.scalar("cost", cost) |
| 36 | + |
| 37 | +with tf.name_scope("train") as scope: |
| 38 | + train = tf.train.GradientDescentOptimizer(learning_rate=0.1).minimize(cost) |
| 39 | + |
| 40 | +# Accuracy computation |
| 41 | +# True if hypothesis>0.5 else False |
| 42 | +predicted = tf.cast(hypothesis > 0.5, dtype=tf.float32) |
| 43 | +accuracy = tf.reduce_mean(tf.cast(tf.equal(predicted, Y), dtype=tf.float32)) |
| 44 | + |
| 45 | +# Launch graph |
| 46 | +with tf.Session() as sess: |
| 47 | + #tensorboard --logdir=./logs/xor_logs |
| 48 | + merged = tf.summary.merge_all() |
| 49 | + writer = tf.summary.FileWriter("./logs/xor_logs", sess.graph) |
| 50 | + |
| 51 | + # Initialize TensorFlow variables |
| 52 | + sess.run(tf.global_variables_initializer()) |
| 53 | + |
| 54 | + for step in range(10001): |
| 55 | + summary, _ = sess.run([merged, train], feed_dict={X: x_data, Y: y_data}) |
| 56 | + if step % 100 == 0: |
| 57 | + print(step, sess.run(cost, feed_dict={ |
| 58 | + X: x_data, Y: y_data}), sess.run([W1, W2])) |
| 59 | + writer.add_summary(summary, step) |
| 60 | + |
| 61 | + # Accuracy report |
| 62 | + h, c, a = sess.run([hypothesis, predicted, accuracy], |
| 63 | + feed_dict={X: x_data, Y: y_data}) |
| 64 | + print("\nHypothesis: ", h, "\nCorrect: ", c, "\nAccuracy: ", a) |
| 65 | + |
| 66 | + |
| 67 | +''' |
| 68 | +Hypothesis: [[ 0.01338218] |
| 69 | + [ 0.98166394] |
| 70 | + [ 0.98809403] |
| 71 | + [ 0.01135799]] |
| 72 | +Correct: [[ 0.] |
| 73 | + [ 1.] |
| 74 | + [ 1.] |
| 75 | + [ 0.]] |
| 76 | +Accuracy: 1.0 |
| 77 | +''' |
0 commit comments