tensorflow学习-第一篇

这两天开始看tensorflow，首先来个神经网络的训练大致过程的python代码

import tensorflow as tf
from numpy.random import RandomState

sess = tf.Session()
batch_size = 8

x = tf.placeholder(tf.float32, shape=(None, 2), name='x-input')
y_ = tf.placeholder(tf.float32, shape=(None, 1), name='y-input')

w1 = tf.Variable(tf.random_normal([2, 1], stddev=1, seed=1))
y = tf.matmul(x, w1)

loss_less = 10
loss_more = 1
loss = tf.reduce_sum(tf.where(tf.greater(y, y_), (y - y_) * loss_more, (y_ - y) * loss_less))
train_step = tf.train.AdamOptimizer(0.001).minimize(loss)

dataset_size = 128
rdm = RandomState(1)
X = rdm.rand(dataset_size, 2)
Y = [[x1 + x2 + rdm.rand() / 10.0-0.05] for (x1, x2) in X]

STEPS = 5000
with tf.Session() as sess:
    init_op = tf.initialize_all_variables()
    sess.run(init_op)
    print(sess.run(w1))
    for i in range(STEPS):
        start = (i * batch_size) % dataset_size
        end = min(start + batch_size, dataset_size)
        sess.run(train_step, feed_dict={x: X[start:end], y_: Y[start:end]})
    print(sess.run(w1))