#===============================================
# 训练简单的神经网络,并显示运行时间
# 数据集：mnnist
#===============================================
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import datetime
starttime = datetime.datetime.now()

import tensorflow as tf
import numpy as np

# Import data
from tensorflow.examples.tutorials.mnist import input_data
flags = tf.app.flags
FLAGS = flags.FLAGS
flags.DEFINE_string('data_dir', '/learn/tensorflow/python/data/', 'Directory for storing data') # 把数据放在/data文件夹中
mnist_data = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)   # 读取数据集

# 建立抽象模型
x = tf.placeholder(tf.float32, [None, 784]) # 占位符
y = tf.placeholder(tf.float32, [None, 10])
W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
a = tf.nn.softmax(tf.matmul(x, W) + b)

# 定义损失函数和训练方法
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y * tf.log(a), reduction_indices=[1]))  # 损失函数为交叉熵，学习速率要设为0.3量级
#cross_entropy = -tf.reduce_sum(y * tf.log(a))   # 损失函数为交叉熵
optimizer = tf.train.GradientDescentOptimizer(0.3) # 梯度下降法，学习速率要设为0.001量级
train_next = optimizer.minimize(cross_entropy) # 训练目标：最小化损失函数

# Train
sess = tf.InteractiveSession()      # 建立交互式会话
# tf.global_variables_initializer().run()
sess.run(tf.global_variables_initializer())
for i in range(1000):
    batch_xs, batch_ys = mnist_data.train.next_batch(100) # 随机抓取100个数据
#    train_next.run({x: batch_xs, y: batch_ys})
    sess.run(train_next, feed_dict={x: batch_xs, y: batch_ys})

#测试
correct_prediction = tf.equal(tf.argmax(a, 1), tf.argmax(y, 1))
# tf.cast先将数据转换成float，防止求平均不准确:比如 tf.float32就是正确，写成tf.float16导致不准确，超出范围。
accuracy = tf.reduce_mean(tf.cast(correct_prediction, 'float'))
print(sess.run(accuracy,feed_dict={x:mnist_data.test.images,y:mnist_data.test.labels}))

endtime=datetime.datetime.now()
print('total time endtime-starttime:', endtime-starttime)