寒假学习日报(四十二)——Tensorflow实验
今天把老师布置的tensorflow实验做了做,由于下载的是2.0版本,不少1.0版本函数调用都需要.compat.v1前缀,还要加入下面这个代码保证运行:
# 保证session.run能够正常运行 tf.compat.v1.disable_eager_execution()
七个实验除了最后一个都已完成:
解决的Bug就是本地MNIST数据集的导入问题,导入函数如下所示:
from tensorflow.examples.tutorials.mnist import input_data mnist = input_data.read_data_sets("MNIST_Data", one_hot=True)
这个是1.0版本的写法,2.0版本开始都使用keras模块了,但keras模块拿到数据后需要手动进行one_hot处理,我不清楚1.0版本中one_hot=True是咋处理的,因此废了一门心思去引入tutorials包。主要处理的BUG是一开始引入MNIST——Data总是报错,最后我把数据集删除后重新下了一个新的,引入成功。。。。
最后就是问题代码:
from __future__ import print_function import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data import os # 保证session.run能够正常运行 tf.compat.v1.disable_eager_execution() os.environ["CUDA_VISIBLE_DEVICES"] = "0" mnist = input_data.read_data_sets("MNIST_Data", one_hot=True) # 设置训练参数 learning_rate = 0.001 training_step = 10000 batch_size = 128 display_step = 400 # 设置双向循环神经网络参数 num_input = 28 timestep = 28 num_hidden = 128 num_classes = 10 # 构造计算图输入变量 X = tf.compat.v1.placeholder("float32", [None, timestep, num_input]) Y = tf.compat.v1.placeholder("float32", [None, num_classes]) # 设置权重和偏值 weights = { 'out': tf.Variable(tf.compat.v1.random_normal([2*num_hidden, num_classes])) } biases = { 'out': tf.Variable(tf.compat.v1.random_normal([num_classes])) } # 自定义双向循环神经网络函数 def BiRNN(X, weights, biases): x = tf.unstack(X, timestep, 1) lstm_fw_cell = tf.compat.v1.nn.rnn_cell.BasicLSTMCell(num_hidden, forget_bias=1.0) lstm_bw_cell = tf.compat.v1.nn.rnn_cell.BasicLSTMCell(num_hidden, forget_bias=1.0) try: outputs, _, _ = tf.compat.v1.nn.bidirectional_dynamic_rnn(lstm_fw_cell, lstm_bw_cell, x, dtype=tf.float32) except Exception: outputs = tf.compat.v1.nn.bidirectional_dynamic_rnn(lstm_fw_cell, lstm_bw_cell, x, dtype=tf.float32) return tf.matmul(outputs[-1], weights['out']) + biases['out'] # 获取神经网络输出层logits,使用softmax激活函数将logits映射成各类取值概率,结果赋值给prediction logits = BiRNN(X, weights, biases) prediction = tf.nn.softmax(logits) # 通过交叉熵构建损失函数,并使用梯度下降法求解,将结果赋值给train_op。 loss_op = tf.reduce_mean(tf.compat.v1.nn.softmax_cross_entropy_with_logits(logits=logits, labels=Y)) optimizer = tf.compat.v1.train.GradientDescentOptimizer(learning_rate=learning_rate) train_op = optimizer.minimize(loss_op) # 计算训练模型的准确率 correct_pred = tf.equal(tf.argmax(prediction, 1), tf.argmax(Y, 1)) accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32)) # 初始化全部变量 init = tf.compat.v1.global_variables_initializer() # Session with tf.compat.v1.Session() as sess: # 训练模型 for step in range(1, training_step+1): batch_x, batch_y = mnist.train.next_batch(batch_size) batch_x = batch_x.reshape((batch_size, timestep, num_input)) sess.run(train_op, feed_dict={X: batch_x, Y: batch_y}) if step % display_step == 0 or step == 1: loss, acc = sess.run([loss_op, accuracy], feed_dict={X: batch_x, Y: batch_y}) print("Step"+str(step)+",Minbatch Loss="+"{:.4f}".format(loss)+",Training Accuracy="+"{:.3f}".format(acc)) print("Optimization Finished!") # 设置测试数据长度为128,通过训练模型对测试数据进行预测,打印测试数据的准确率 test_len = 128 test_data = mnist.test.images[:test_len].reshape((-1, timestep, num_input)) test_label = mnist.test.labels["test_len"] print("Test Accuracy:", sess.run(accuracy, feed_dict={X: test_data, Y: test_label}))
主要报ValueError错误,目前还在排查中。。。。