Tensorflow Manage Experiments

1. Visualize graphs with TensorBoard

# define model
# launch a session to compute the graph
with tf.Session() as sess:
    writer = tf.summary.FileWriter("./graphs", sess.graph)       #
    for step in range(training_steps):
        sess.run([optimizer])
# Go to terminal, run:
# $ python [yourprogram].py
# $ tensorboard --logdir="./graphs" --port 6006
# Then open your browser and go to: http://localhost:6006/

2. Saving and Restoring Variables

# checkpoint
global_step = tf.Variable(0, dtype=tf.int32, trainable=False, name="global_step")  # 
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy,
                                                   global_step = global_step)   # 
# define model
# launch a session to compute the graph
saver = tf.train.Saver()    # 
with tf.Session() as sess:
    for step in range(training_steps):
        sess.run([train_op])
        if (i + 1)% 300 == 0:  # 
            saver.save(sess, './checkpoints/ckpt', global_step=global_step)  # 

# define model
# launch a session to compute the graph
with tf.Session() as sess:
    ckpt = tf.train.get_checkpoint_state(os.path.dirname('./checkpoints/ckpt'))   # 
    if ckpt and ckpt.model_checkpoint_path:    # 
        saver.restore(sess, ckpt.model_checkpoint_path)   # 
    for step in range(training_steps):
        sess.run([optimizer])   

3. Visualize our summary statistics during our training

# define model
with tf.name_scope("summaries"):   #      
    tf.summary.image('input', x_image, 4) #  
    tf.summary.scalar("accuracy", accuracy)  #  
    tf.summary.histogram("loss", cross_entropy)#   
    summary_op = tf.summary.merge_all()  #  

# launch a session to compute the graph
with tf.Session() as sess:    
    writer = tf.summary.FileWriter("./graphs", sess.graph)    #   
    for step in range(training_steps):
        sess.run([train_op])
        summary = sess.run(summary_op, feed_dict={x: batch[0], y_: batch[1], 
                                                  keep_prob: 0.5})       #   
        writer.add_summary(summary, global_step = i)     #