NN_basic_Classification.py

import tensorflow as tf
import numpy as np

x_data = np.array([[0,0],[1,0],[1,1],[0,0],[0,0],[0,1]])

y_data = np.array([
    [1,0,0],
    [0,1,0],
    [0,0,1],
    [1,0,0],
    [1,0,0],
    [0,0,1]])

X = tf.placeholder(tf.float32)
Y = tf.placeholder(tf.float32)

W = tf.Variable(tf.random_uniform([2,3],-1.,1.))


b = tf.Variable(tf.zeros([3]))

L = tf.add(tf.matmul(X,W),b)
L = tf.nn.relu(L)

model = tf.nn.softmax(L)

cost = tf.reduce_mean(- tf.reduce_sum(Y * tf.log(model),axis=1))

opt = tf.train.GradientDescentOptimizer(learning_rate=0.01)

train_opt = opt.minimize(cost)


init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

for step in range(100):
    sess.run(train_opt, feed_dict={X: x_data, Y: y_data})

    if (step + 1) % 10 == 0:
        print(step + 1, sess.run(cost, feed_dict={X : x_data, Y :y_data}))



prediction = tf.argmax(model,1)
target = tf.argmax(Y,1)

print('예측값 : ', sess.run(prediction, feed_dict={X : x_data}))
print('실제값 : ', sess.run(target, feed_dict={Y : y_data}))

is_correct = tf.equal(prediction, target)
accurary = tf.reduce_mean(tf.cast(is_correct,tf.float32))
print('정확도 : %.2f' % sess.run(accurary*100, feed_dict={X : x_data,Y : y_data}))