tensorboard可视化

  1 #*- coding:utf-8 -*
  2 # https://www.cnblogs.com/felixwang2/p/9184404.html
  3 # TensorFlow(八)：tensorboard可视化
  4 
  5 import tensorflow as tf
  6 from tensorflow.examples.tutorials.mnist import input_data
  7 from tensorflow.contrib.tensorboard.plugins import projector
  8 
  9 # 载入数据集
 10 mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
 11 # 运行次数
 12 max_steps = 1001
 13 # 图片数量
 14 image_num = 3000  # 最多10000，因为测试集为10000
 15 # 文件路径
 16 DIR = "F:/document/PyCharm/temp/"
 17 
 18 # 定义会话
 19 gpu_options = tf.GPUOptions(allow_growth=True)
 20 sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
 21 
 22 # 载入图片
 23 embedding = tf.Variable(tf.stack(mnist.test.images[:image_num]), trainable=False, name=‘embedding‘)
 24 
 25 
 26 # 参数概要
 27 def variable_summaries(var):
 28     with tf.name_scope(‘summaries‘):
 29         mean = tf.reduce_mean(var)
 30         tf.summary.scalar(‘mean‘, mean)  # 平均值
 31         with tf.name_scope(‘stddev‘):
 32             stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
 33         tf.summary.scalar(‘stddev‘, stddev)  # 标准差
 34         tf.summary.scalar(‘max‘, tf.reduce_max(var))  # 最大值
 35         tf.summary.scalar(‘min‘, tf.reduce_min(var))  # 最小值
 36         tf.summary.histogram(‘histogram‘, var)  # 直方图
 37 
 38 
 39 # 命名空间
 40 with tf.name_scope(‘input‘):
 41     # 这里的none表示第一个维度可以是任意的长度
 42     x = tf.placeholder(tf.float32, [None, 784], name=‘x-input‘)
 43     # 正确的标签
 44     y = tf.placeholder(tf.float32, [None, 10], name=‘y-input‘)
 45 
 46 # 显示图片
 47 with tf.name_scope(‘input_reshape‘):
 48     image_shaped_input = tf.reshape(x, [-1, 28, 28, 1])  # -1表示不确定的值
 49     tf.summary.image(‘input‘, image_shaped_input, 10)  # 一共放10张图片
 50 
 51 with tf.name_scope(‘layer‘):
 52     # 创建一个简单神经网络
 53     with tf.name_scope(‘weights‘):
 54         W = tf.Variable(tf.zeros([784, 10]), name=‘W‘)
 55         variable_summaries(W)
 56     with tf.name_scope(‘biases‘):
 57         b = tf.Variable(tf.zeros([10]), name=‘b‘)
 58         variable_summaries(b)
 59     with tf.name_scope(‘wx_plus_b‘):
 60         wx_plus_b = tf.matmul(x, W) + b
 61     with tf.name_scope(‘softmax‘):
 62         prediction = tf.nn.softmax(wx_plus_b)
 63 
 64 with tf.name_scope(‘loss‘):
 65     # 交叉熵代价函数
 66     loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y, logits=prediction))
 67     tf.summary.scalar(‘loss‘, loss)
 68 with tf.name_scope(‘train‘):
 69     # 使用梯度下降法
 70     train_step = tf.train.GradientDescentOptimizer(0.5).minimize(loss)
 71 
 72 # 初始化变量
 73 sess.run(tf.global_variables_initializer())
 74 
 75 with tf.name_scope(‘accuracy‘):
 76     with tf.name_scope(‘correct_prediction‘):
 77         # 结果存放在一个布尔型列表中
 78         correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction, 1))  # argmax返回一维张量中最大的值所在的位置
 79     with tf.name_scope(‘accuracy‘):
 80         # 求准确率
 81         accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))  # 把correct_prediction变为float32类型
 82         tf.summary.scalar(‘accuracy‘, accuracy)
 83 
 84 # 产生metadata文件
 85 if tf.gfile.Exists(DIR + ‘projector/projector/metadata.tsv‘):  # 检测是否已存在
 86     tf.gfile.DeleteRecursively(DIR + ‘projector/projector/metadata.tsv‘)
 87 with open(DIR + ‘projector/projector/metadata.tsv‘, ‘w‘) as f:
 88     labels = sess.run(tf.argmax(mnist.test.labels[:], 1))
 89     for i in range(image_num):
 90         f.write(str(labels[i]) + ‘\n‘)
 91 
 92         # 合并所有的summary
 93 merged = tf.summary.merge_all()
 94 
 95 projector_writer = tf.summary.FileWriter(DIR + ‘projector/projector‘, sess.graph)
 96 saver = tf.train.Saver()  # 用来保存网络模型
 97 config = projector.ProjectorConfig()  # 定义了配置文件
 98 embed = config.embeddings.add()
 99 embed.tensor_name = embedding.name
100 embed.metadata_path = DIR + ‘projector/projector/metadata.tsv‘
101 embed.sprite.image_path = DIR + ‘projector/data/mnist_10k_sprite.png‘
102 embed.sprite.single_image_dim.extend([28, 28])
103 projector.visualize_embeddings(projector_writer, config)  # 可视化的一个工具
104 
105 for i in range(max_steps):
106     # 每个批次100个样本
107     batch_xs, batch_ys = mnist.train.next_batch(100)
108 
109     run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
110     run_metadata = tf.RunMetadata()
111 
112     summary, _ = sess.run([merged, train_step], feed_dict={x: batch_xs, y: batch_ys}, options=run_options,
113                           run_metadata=run_metadata)
114     projector_writer.add_run_metadata(run_metadata, ‘step%03d‘ % i)
115     projector_writer.add_summary(summary, i)
116 
117     # 每训练100次打印准确率
118     if i % 100 == 0:
119         acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})
120         print("Iter " + str(i) + ", Testing Accuracy= " + str(acc))
121 
122 # 训练完保存模型
123 saver.save(sess, DIR + ‘projector/projector/a_model.ckpt‘, global_step=max_steps)
124 projector_writer.close()
125 sess.close()