=======数据说明=======
text1.txt
0.10, 0.21, 0.56, 0.72, 0.31
0.23, 0.34, 0.3, 0.76, 0.21
text2.txt
0.60, 0.92, 0.12, 0.83, 0.11
0.89, 0.71, 0.90, 0.40, 0.55
======================
#搭建输入管道 import tensorflow as tf import numpy as np graph = tf.Graph() session = tf.compat.v1.InteractiveSession(graph = graph)
1）文件读取
filenames = [‘text%d.txt‘%i for i in range(1,3)] filename_queue = tf.compat.v1.train.string_input_producer(filenames,capacity=2,shuffle=True,name=‘string_input_producer‘) reader = tf.compat.v1.TextLineReader() _,value = reader.read(filename_queue,name=‘text_read‘)
根据文件名生产文件名队列filename_queue
读取器类reader将文件队列作为参数。实验数据中一行为一个数据，因此使用TextLineReader
value返回读取器从每一行读到的值
2）文本解码
record_defaults = [[-1.0]]*5
col1,col2,col3,col4,col5 = tf.compat.v1.decode_csv(value,record_defaults = record_defaults)
record_default，指定分割后每个属性的类型，也可以不指定类型[]
decode_csv,value原始是string，解析为数字
3）转成训练格式
features = tf.stack([col1,col2,col3,col4,col5]) x = tf.compat.v1.train.shuffle_batch([features],batch_size=3,capacity=5,name=‘batch‘,min_after_dequeue=1,num_threads=1)
stack方法将各个列拼一个张量
shuffle_batch将张量打乱批量输出；train.batch不打乱数据输出
4）线程管理
coord = tf.train.Coordinator()
threads = tf.compat.v1.train.start_queue_runners(coord = coord,sess = session)
任务结束后需要显示的停止线程，加入主线程
coord.request_stop()
coord.join(threads)
5）作为模型输入
W = tf.compat.v1.Variable(tf.compat.v1.random_uniform(shape = [5,5], minval = -0.1, maxval = 0.1, dtype = tf.float32),name = ‘W‘) b = tf.compat.v1.Variable(tf.zeros(shape = [5], dtype = tf.float32), name = ‘b‘) #variable 变量需要初始化random_uniform在minval-maxval之间均匀采样 h = tf.nn.sigmoid(tf.matmul(x,W)+b) tf.compat.v1.global_variables_initializer().run() for step in range(3): x_eval,h_eval = session.run([x,h]) print(‘==========Step%d===========‘%step) print(‘data(x)‘) print(x_eval) print(‘data(h)‘) print(h_eval)
==========Step0=========== Evaluated data(x) [[0.6 0.92 0.12 0.83 0.11] [0.1 0.21 0.56 0.72 0.31] [0.23 0.34 0.3 0.76 0.21]] Evaluated data(h) [[0.5236712 0.5169494 0.4832768 0.49572295 0.50606287] [0.5182559 0.51618594 0.48345992 0.5071334 0.499376 ] [0.5158963 0.51570374 0.4850265 0.50364953 0.50234705]] ==========Step1=========== Evaluated data(x) [[0.89 0.71 0.9 0.4 0.55] [0.6 0.92 0.12 0.83 0.11] [0.23 0.34 0.3 0.76 0.21]] Evaluated data(h) [[0.5350291 0.5160627 0.48820648 0.49639583 0.50967765] [0.5236712 0.5169494 0.4832768 0.49572295 0.50606287] [0.5158963 0.51570374 0.4850265 0.50364953 0.50234705]] ==========Step2=========== Evaluated data(x) [[0.1 0.21 0.56 0.72 0.31] [0.89 0.71 0.9 0.4 0.55] [0.6 0.92 0.12 0.83 0.11]] Evaluated data(h) [[0.5182559 0.51618594 0.48345992 0.5071334 0.499376 ] [0.5350291 0.5160627 0.48820648 0.49639583 0.50967765] [0.5236712 0.5169494 0.4832768 0.49572295 0.50606287]]
6）采坑指南
出现如下报错一般为数据本身出错，如某行数据列数不符合要求
RandomShuffleQueue ‘_55_batch/random_shuffle_queue‘ is closed and has insufficient elements (requested 3, current size 2)
原文：https://www.cnblogs.com/jessyswing/p/12920396.html