tensorflow学习——tf.train.Supervisor()与tf.train.saver()

约定不等于承诺〃 2022-03-15 04:46 279阅读 0赞

原文：[https://blog.csdn.net/ei1990/article/details/77948001][https_blog.csdn.net_ei1990_article_details_77948001]

1、tf.train.Supervisor()

import tensorflow as tf
    import numpy as np
    import os
    
    log_path = 'ckptdir/'
    log_name = 'liner.ckpt'
    x_data = np.random.rand(100).astype(np.float32)
    y_data = x_data*0.1 + 0.3
    
    w = tf.Variable(tf.random_normal([1]))
    b = tf.Variable(tf.zeros([1]))
    y = w*x_data + b
    
    loss = tf.reduce_mean(tf.square(y-y_data))
    train = tf.train.AdamOptimizer(0.5).minimize(loss)
    
    tf.summary.scalar('loss', loss)
    
    saver = tf.train.Saver()
    init = tf.global_variables_initializer()
    merged = tf.summary.merge_all()
    
    sv = tf.train.Supervisor(logdir=log_path, init_op=init)  # logdir用来保存checkpoint和summary
    saver = sv.saver   # 创建saver
    
    with sv.managed_session() as sess:  # 会自动去logdir中去找checkpoint，如果没有的话，自动执行初始化
    #    sess.run(init)
    #    if len(os.listdir(log_path)) != 0:
    #        saver.restore(sess, os.path.join(log_path, log_name))
        for step in range(201):
            sess.run(train)
            if step%50 == 0:
                print(step, sess.run(w), sess.run(b))
                merged_summary = sess.run(merged)
                sv.summary_computed(sess, merged_summary,global_step=step)
        saver.save(sess, os.path.join(log_path, 'liner.ckpt'))

从上面代码可以看出，Supervisor帮助我们处理一些事情  
（1）自动去checkpoint加载数据或初始化数据  
（2）自身有一个Saver，可以用来保存checkpoint  
（3）有一个summary\_computed用来保存Summary  
所以，我们就不需要：  
（1）手动初始化或从checkpoint中加载数据  
（2）不需要创建Saver，使用sv内部的就可以  
（3）不需要创建summary writer  
2、tf.train.Saver()

import tensorflow as tf
    import numpy as np
    import os
    
    log_path = 'ckptdir'
    log_name = 'liner.ckpt'
    x_data = np.random.rand(100).astype(np.float32)
    y_data = x_data*0.1 + 0.3
    
    w = tf.Variable(tf.random_normal([1]))
    b = tf.Variable(tf.zeros([1]))
    y = w*x_data + b
    
    loss = tf.reduce_mean(tf.square(y-y_data))
    train = tf.train.AdamOptimizer(0.5).minimize(loss)
    tf.summary.scalar('loss', loss)
    
    saver = tf.train.Saver()
    init = tf.global_variables_initializer()
    merged = tf.summary.merge_all()
    
    with tf.Session() as sess:
        sess.run(init)
        print("loading model from checkpoint")
        checkpoint = tf.train.latest_checkpoint(os.path.join(log_path, log_name))
        restore_saver.restore(sess, checkpoint)
        #if len(os.listdir(log_path)) != 0:
        #    saver.restore(sess, os.path.join(log_path, log_name))
        for step in range(201):
            sess.run(train)
            if step%50 ==0:
                print(step, sess.run(w), sess.run(b))
        summary_writer = tf.summary.FileWriter(log_path, sess.graph)
        summary_all = sess.run(merged)
        summary_writer.add_summary(summary_all)
        summary_writer.close()
    
        saver.save(sess, os.path.join(log_path, 'liner.ckpt'))

[https_blog.csdn.net_ei1990_article_details_77948001]: https://blog.csdn.net/ei1990/article/details/77948001