XOR을 신경망으로 분류해보자
XOR 데이터
Logistic Regression 모델
import numpy as np
import tensorflow as tf
x_data = np.array([[0, 0],
[0, 1],
[1, 0],
[1, 1]], dtype=np.float32)
y_data = np.array([[0],
[1],
[1], [0]], dtype=np.float32)
X = tf.placeholder(tf.float32)
Y = tf.placeholder(tf.float32)
W = tf.Variable(tf.random_normal([2, 1]), name='weight')
b = tf.Variable(tf.random_normal([1]), name='bias')
hypothesis = tf.sigmoid(tf.matmul(X, W) + b)
cost = -tf.reduce_mean(Y * tf.log(hypothesis) + (1 - Y) * tf.log(1 - hypothesis))
train = tf.train.GradientDescentOptimizer(learning_rate=0.1).minimize(cost)
# 모델의 정확도 계산
# hypothesis > 0 이면 참, 아니면 거짓)
predicted = tf.cast(hypothesis > 0.5, dtype=tf.float32)
accuracy = tf.reduce_mean(tf.cast(tf.equal(predicted, Y), dtype=tf.float32))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for step in range(10001):
sess.run(train, feed_dict={X: x_data, Y: y_data})
if step % 1000 == 0:
print(f'step = {step:06}, cost : {sess.run(cost, feed_dict={X: x_data, Y: y_data})}, '
f'weight : {tf.squeeze(sess.run(W)).eval(session=sess)}')
h, c, a = sess.run([hypothesis, predicted, accuracy], feed_dict={X: x_data, Y: y_data})
print(f'Hypothesis : {tf.squeeze(h).eval(session=sess)}')
print(f'Correct : {tf.squeeze(c).eval(session=sess)}')
print(f'Accuracy : {a*100:.6}%')step = 000000, cost : 0.8311277627944946, weight : [-0.59255368 2.05449486]
step = 001000, cost : 0.6931628584861755, weight : [ 0.01237938 0.01779066]
step = 002000, cost : 0.6931471824645996, weight : [ 0.00029095 0.0003012 ]
step = 003000, cost : 0.6931471824645996, weight : [ 5.78627623e-06 5.79754351e-06]
step = 004000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
step = 005000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
step = 006000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
step = 007000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
step = 008000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
step = 009000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
step = 010000, cost : 0.6931471824645996, weight : [ 1.32766417e-07 1.32112120e-07]
Hypothesis : [ 0.5 0.5 0.5 0.5]
Correct : [ 0. 0. 0. 0.]
Accuracy : 50.0%
Neural Network 모델
import numpy as np
import tensorflow as tf
x_data = np.array([[0, 0],
[0, 1],
[1, 0],
[1, 1]], dtype=np.float32)
y_data = np.array([[0],
[1],
[1], [0]], dtype=np.float32)
X = tf.placeholder(tf.float32)
Y = tf.placeholder(tf.float32)
# Neural Network 적용
W1 = tf.Variable(tf.random_normal([2, 1]), name='weight1')
b1 = tf.Variable(tf.random_normal([2]), name='bias1')
layer1 = tf.sigmoid(tf.matmul(X, W1) + b1)
W2 = tf.Variable(tf.random_normal([2, 1]), name='weight2')
b2 = tf.Variable(tf.random_normal([1]), name='bias2')
hypothesis = tf.sigmoid(tf.matmul(layer1, W2) + b2)
cost = -tf.reduce_mean(Y * tf.log(hypothesis) + (1 - Y) * tf.log(1 - hypothesis))
train = tf.train.GradientDescentOptimizer(learning_rate=0.1).minimize(cost)
# 모델의 정확도 계산
# hypothesis > 0 이면 참, 아니면 거짓)
predicted = tf.cast(hypothesis > 0.5, dtype=tf.float32)
accuracy = tf.reduce_mean(tf.cast(tf.equal(predicted, Y), dtype=tf.float32))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for step in range(10001):
sess.run(train, feed_dict={X: x_data, Y: y_data})
if step % 1000 == 0:
print(f'step = {step:06}, cost : {sess.run(cost, feed_dict={X: x_data, Y: y_data}):.10}, '
f'weight : {tf.squeeze(sess.run(W1)).eval(session=sess)}, '
f'{tf.squeeze(sess.run(W2)).eval(session=sess)}')
h, c, a = sess.run([hypothesis, predicted, accuracy], feed_dict={X: x_data, Y: y_data})
print(f'Hypothesis : {tf.squeeze(h).eval(session=sess)}')
print(f'Correct : {tf.squeeze(c).eval(session=sess)}')
print(f'Accuracy : {a*100:.6}%')step = 000000, cost : 0.6963943839, weight : [ 0.37977377 -0.03527401], [-0.2952694 -0.10899331]
step = 001000, cost : 0.6927730441, weight : [ 0.42299902 -0.20793314], [-0.39537558 -0.13621624]
step = 002000, cost : 0.6917776465, weight : [ 0.57296711 -0.44276038], [-0.50253206 -0.09007944]
step = 003000, cost : 0.6853944063, weight : [ 0.98696107 -0.93629724], [-0.83750314 0.06653196]
step = 004000, cost : 0.4990587234, weight : [ 2.54613757 -2.89969301], [-2.36907244 1.84388542]
step = 005000, cost : 0.1204189509, weight : [ 4.94409466 -5.14697409], [-5.11362267 5.43273211]
step = 006000, cost : 0.05705790967, weight : [ 5.75085926 -5.89725971], [-6.50368214 6.9214716 ]
step = 007000, cost : 0.03641526401, weight : [ 6.16756582 -6.2878871 ], [-7.34480333 7.78554535]
step = 008000, cost : 0.02652502432, weight : [ 6.43835545 -6.54318571], [-7.94078159 8.3894968 ]
step = 009000, cost : 0.02078646049, weight : [ 6.63547897 -6.72984171], [-8.4005928 8.85260773]
step = 010000, cost : 0.01705784164, weight : [ 6.78890705 -6.87561798], [-8.77430153 9.2277565 ]
Hypothesis : [ 0.01516829 0.97779286 0.98493963 0.01519784]
Correct : [ 0. 1. 1. 0.]
Accuracy : 100.0%
Neural Network에서 neural(=weight)를 늘리려면?
W1 = tf.Variable(tf.random_normal([2, 10]), name='weight1')
b1 = tf.Variable(tf.random_normal([10]), name='bias1')
layer1 = tf.sigmoid(tf.matmul(X, W1) + b1)
W2 = tf.Variable(tf.random_normal([10, 1]), name='weight2')
b2 = tf.Variable(tf.random_normal([1]), name='bias2')
hypothesis = tf.sigmoid(tf.matmul(layer1, W2) + b2)
Hypothesis : [ 0.00381555 0.99342114 0.99458128 0.00839256]
Correct : [ 0. 1. 1. 0.]
Accuracy : 100.0%
Deep Neural Network
W1 = tf.Variable(tf.random_normal([2, 10]), name='weight1')
b1 = tf.Variable(tf.random_normal([10]), name='bias1')
layer1 = tf.sigmoid(tf.matmul(X, W1) + b1)
W2 = tf.Variable(tf.random_normal([10, 10]), name='weight2')
b2 = tf.Variable(tf.random_normal([10]), name='bias2')
layer2 = tf.sigmoid(tf.matmul(layer1, W2) + b2)
W3 = tf.Variable(tf.random_normal([10, 10]), name='weight2')
b3 = tf.Variable(tf.random_normal([10]), name='bias2')
layer3 = tf.sigmoid(tf.matmul(layer2, W3) + b3)
W4 = tf.Variable(tf.random_normal([10, 1]), name='weight2')
b4 = tf.Variable(tf.random_normal([1]), name='bias2')
hypothesis = tf.sigmoid(tf.matmul(layer3, W4) + b4)
Hypothesis : [ 0.0014578 0.99773043 0.99896216 0.00163728]
Correct : [ 0. 1. 1. 0.]
Accuracy : 100.0%
