Attempting to use uninitialized value input_producer/limit_epochs/epochs

Tensowflow语言类似于Spark Streaming或者storm　DSL语言，先定义computer graph，最后run。调试困难。直接上代码

import tensorflow as tf

with tf.Session() as sess:

    filename_queue = tf.train.string_input_producer(["file0.csv", "file1.csv", "file2.csv"],num_epochs = 5)

    #sess.run(tf.initialize_all_variables())

    #sess.run(tf.initialize_local_variables())

    reader = tf.WholeFileReader()

    key, value = reader.read(filename_queue)

    threads = tf.train.start_queue_runners(sess=sess)

    for _ in range(10):

        k, v = sess.run([key, value])

        print("key:{},value:{}".format(k,v))

num_epochs居然被定义为local variable。得先initialize。Tensowflow的 variables initilize特别的坑。有如下几种：

1.tf.initialize_all_variables()，不处理local variable

2.tf.initialize_local_variables()，只处理local variable。

最大的问题，类似于声明式语言如C，变量得先定义，然后才能赋值。本段代码中epochs是在下面语句中声明，

 filename_queue = tf.train.string_input_producer(["file0.csv", "file1.csv", "file2.csv"],num_epochs = 5)，

赋值只能放在这条语句下面!!!!!!!!!!!!!!!!!!!!!!否则还是uninitialized，狗屎