基于TensorFlow进行TensorBoard可视化

  1 # -*- coding: utf-8 -*-
  2 """
  3 Created on Thu Nov  1 17:51:28 2018
  4
  5 @author: zhen
  6 """
  7
  8 import tensorflow as tf
  9 from tensorflow.examples.tutorials.mnist import input_data
 10
 11 max_steps = 1000
 12 learning_rate = 0.001
 13 dropout = 0.9
 14 data_dir = ‘C:/Users/zhen/MNIST_data_bak/‘
 15 log_dir = ‘C:/Users/zhen/MNIST_log_bak/‘
 16
 17 mnist = input_data.read_data_sets(data_dir, one_hot=True)
 18 sess = tf.InteractiveSession()
 19
 20 with tf.name_scope(‘input‘):
 21     x = tf.placeholder(tf.float32, [None, 784], name=‘x-inpupt‘)
 22     y_ = tf.placeholder(tf.float32, [None, 10], name=‘y-input‘)
 23
 24 with tf.name_scope("input_reshape"):
 25     image_shaped_input = tf.reshape(x, [-1, 28, 28, 1])
 26     tf.summary.image(‘input‘, image_shaped_input, 10)
 27
 28 # 定义神经网络的初始化方法
 29 def weight_variable(shape):
 30     initial = tf.truncated_normal(shape, stddev=0.1)
 31     return tf.Variable(initial)
 32
 33 def bias_variable(shape):
 34     initial = tf.constant(0.1, shape=shape)
 35     return tf.Variable(initial)
 36
 37 # 定义Variable变量的数据汇总函数
 38 def variable_summaries(var):
 39     with tf.name_scope(‘summaries‘):
 40         mean = tf.reduce_mean(var)
 41         tf.summary.scalar(‘mean‘, mean)
 42         with tf.name_scope(‘stddev‘):
 43             stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
 44         tf.summary.scalar(‘stddev‘, stddev)
 45         tf.summary.scalar(‘max‘, tf.reduce_max(var))
 46         tf.summary.scalar(‘min‘, tf.reduce_min(var))
 47         tf.summary.histogram(‘histogram‘, var)
 48
 49 # 创建MLP多层神经网络
 50 def nn_layer(input_tensor, input_dim, output_dim, layer_name, act=tf.nn.relu):
 51     with tf.name_scope(layer_name):
 52         with tf.name_scope(‘weights‘):
 53             weights = weight_variable([input_dim, output_dim])
 54             variable_summaries(weights)
 55         with tf.name_scope(‘biases‘):
 56             biases = bias_variable([output_dim])
 57             variable_summaries(biases)
 58         with tf.name_scope(‘Wx_plus_b‘):
 59             preactivate = tf.matmul(input_tensor, weights) + biases
 60             tf.summary.histogram(‘pre_activations‘, preactivate)
 61         activations = act(preactivate, name=‘activation‘)
 62         tf.summary.histogram(‘activations‘, activations)
 63         return activations
 64
 65 hidden1 = nn_layer(x, 784, 500, ‘layer1‘)
 66
 67 with tf.name_scope(‘dropout‘):
 68     keep_prob = tf.placeholder(tf.float32)
 69     tf.summary.scalar(‘dropout_keep_probability‘, keep_prob)
 70     dropped  = tf.nn.dropout(hidden1, keep_prob)
 71
 72 y = nn_layer(dropped, 500, 10, ‘layer2‘, act=tf.identity)
 73
 74 with tf.name_scope(‘scross_entropy‘):
 75     diff = tf.nn.softmax_cross_entropy_with_logits(logits=y, labels=y_)
 76     with tf.name_scope(‘total‘):
 77         cross_entropy  = tf.reduce_mean(diff)
 78 tf.summary.scalar(‘cross_entropy‘, cross_entropy)
 79
 80 with tf.name_scope(‘train‘):
 81     train_step = tf.train.AdamOptimizer(learning_rate).minimize(cross_entropy)
 82 with tf.name_scope(‘accuracy‘):
 83     with tf.name_scope(‘accuracy‘):
 84         correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
 85     with tf.name_scope(‘accuracy‘):
 86         accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
 87
 88 tf.summary.scalar(‘accuracy‘, accuracy)
 89
 90 merged = tf.summary.merge_all()
 91 train_writer = tf.summary.FileWriter(log_dir + ‘/train‘, sess.graph)
 92 test_writer = tf.summary.FileWriter(log_dir + ‘/test‘) # 可视化数据存储在日志文件中
 93 tf.global_variables_initializer().run()
 94
 95 def feed_dict(train):
 96     if train:
 97         xs, ys = mnist.train.next_batch(100)
 98         k = dropout
 99     else:
100         xs, ys = mnist.test.images, mnist.test.labels
101         k = 1.0
102     return {x:xs, y_:ys, keep_prob:k}
103
104 saver = tf.train.Saver()
105 for i in range(max_steps):
106     if i % 100 == 0:
107         summary, acc = sess.run([merged, accuracy], feed_dict(False))
108         test_writer.add_summary(summary, i)
109         print(‘Accuray at step %s:%s‘ % (i, acc))
110     else:
111         if i % 100 == 99:
112             run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
113             run_metadata = tf.RunMetadata()
114             summary, _ = sess.run([merged, train_step], feed_dict=feed_dict(True))
115             train_writer.add_run_metadata(run_metadata, ‘stp%03d‘ % i)
116             train_writer.add_summary(summary, 1)
117             saver.save(sess, log_dir + ‘model.ckpt‘, i)
118         else:
119             summary, _ = sess.run([merged,train_step], feed_dict=feed_dict(True))
120             train_writer.add_summary(summary, i)
121
122 train_writer.close()
123 test_writer.close()
124
125     

　　程序执行完成后，在dos命令窗口或linux窗口运行命令，参数logdir是你程序保存log日志设置的地址。

效果如下：

结果：

　　一层神经网络：

　　二层神经网络：

　　神经网络计算图：

　　神经元输出的分布：

　　数据分布直方图：
　　　　　　

　　数据可视化：

原文地址：https://www.cnblogs.com/yszd/p/9910456.html