10分钟搞懂TensorBoard用法

基本用法

启动采集器，将运行session环境内的参数都保存到文件里，后续就可以用

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    file_writer = tf.summary.FileWriter('./logs/1', sess.graph)

后续通过TensorBoard打开这个文件，查看这个session的模型，运行

tensorboard --logdir=./logs/1

打开浏览器，通常是通过本机的6006端口访问

tensorboard

对模型归类

在session中，对模型做归类

 with tf.name_scope("RNN_layers”):
        lstm = tf.contrib.rnn.BasicLSTMCell(lstm_size)
        drop = tf.contrib.rnn.DropoutWrapper(lstm, output_keep_prob=keep_prob)
        cell = tf.contrib.rnn.MultiRNNCell([drop] * num_layers)

with tf.name_scope(“RNN_layers”)

查看session文件效果如图

tensorboard name space

采集运行信息

 with tf.name_scope('logits’):
        softmax_w = tf.Variable(tf.truncated_normal((lstm_size, num_classes), stddev=0.1),
                               name=‘softmax_w’)
        softmax_b = tf.Variable(tf.zeros(num_classes), name=‘softmax_b’)
        logits = tf.matmul(output, softmax_w) + softmax_b
        tf.summary.histogram('softmax_w', softmax_w)
        tf.summary.histogram('softmax_b', softmax_b) #以直方图采集权重 
….
merged = tf.summary.merge_all() #收集全部采集点
…..
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    train_writer = tf.summary.FileWriter('./logs/2/train', sess.graph)
    ….
    summary, batch_loss, new_state, _ = sess.run([model.merged, model.cost, 
                                                          model.final_state, model.optimizer], 
                                                          feed_dict=feed) #运行采集点的收集器merge
    ….
    train_writer.add_summary(summary, iteration) #采集到的信息写入文件

tensorboard histogram

比较模型的不同参数，调参用

epochs = 20
batch_size = 100
num_steps = 100
train_x, train_y, val_x, val_y = split_data(chars, batch_size, num_steps)

for lstm_size in [128,256,512]:
    for num_layers in [1, 2]:
        for learning_rate in [0.002, 0.001]:
            log_string = 'logs/4/lr={},rl={},ru={}'.format(learning_rate, num_layers, lstm_size) #每一对参数写入一个文件
            writer = tf.summary.FileWriter(log_string)
            model = build_rnn(len(vocab), 
                    batch_size=batch_size,
                    num_steps=num_steps,
                    learning_rate=learning_rate,
                    lstm_size=lstm_size,
                    num_layers=num_layers)
            
            train(model, epochs, writer)#每个文件用采集器收集信息

对每个参数配置做记录，最终可以得到他们之间对比的图案

image

总结：

• 图形做归类： with tf.name_scope('logits’):
• 埋点：tf.summary.histogram('softmax_w', softmax_w)
• 打包点：merged = tf.summary.merge_all()
• 设置读写文件： train_writer = tf.summary.FileWriter('./logs/2/train', sess.graph)
• 运行时做记录：train_writer.add_summary(summary, iteration)

关于我：

linxinzhe，全栈工程师，目前供职于某世界500强银行的金融科技部门（人工智能，区块链）。

GitHub:https://github.com/linxinzhe

欢迎留言讨论，也欢迎关注我~
我也会关注你的哦！