tensorboard 使用教程

---

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

mnist=input_data.read_data_sets(r"D:\Jupyter\MNIST-data",one_hot=True)

def get_weight(shape,name):
    return tf.Variable(tf.truncated_normal(shape,stddev=0.1),name=name)

def get_bias(shape,name):
    return tf.Variable(tf.zeros(shape),name=name)

def conv2d(X,W):
    return tf.nn.conv2d(X,W,strides=[1,1,1,1],padding="SAME")
def pool2d(X):
    return tf.nn.max_pool(X,ksize=[1,2,2,1],strides=[1,2,2,1],padding="SAME")

with tf.name_scope("input"):
    x=tf.placeholder('float',shape=[None,784],name="x")
    y=tf.placeholder('float',shape=[None,10],name="y")
    x_images=tf.reshape(x,[-1,28,28,1],name="x_images")
    tf.summary.image("pic",x_images,9)

with tf.name_scope("conv1"):
    W1=get_weight([5,5,1,64],"W1")
    tf.summary.histogram("W1",W1)
    b1=get_bias([64],"b1")
    tf.summary.histogram("b1",b1)
    conv1=tf.nn.relu(conv2d(x_images,W1)+b1,name="conv1")
    pool1=pool2d(conv1)

with tf.name_scope("conv2"):
    W2=get_weight([3,3,64,32],"W2")
    b2=get_bias([32],"b2")
    conv2=tf.nn.relu(conv2d(pool1,W2)+b2,name="conv2")
    pool2=pool2d(conv2)

with tf.name_scope("FC1"):
    flatten=tf.reshape(pool2,[-1,7*7*32],name="flatten")
    W3=get_weight([7*7*32,128],"W3")
    b3=get_bias([128],"b3")
    FC1=tf.nn.relu(tf.matmul(flatten,W3)+b3,name="FC1")

with tf.name_scope("FC2"):
    W4=get_weight([128,10],"W4")
    b4=get_bias([10],"b4")
    out=tf.add(tf.matmul(FC1,W4),b4,name="out")

with tf.name_scope("loss"):
    loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y,logits=out),name="loss")
    tf.summary.scalar("loss",loss)
    tf.summary.histogram("loss",loss)
with tf.name_scope("train"):
    optimizer=tf.train.AdamOptimizer(0.001)
    train_step=optimizer.minimize(loss,name="train")

with tf.name_scope("acc"):
    pre_correct=tf.equal(tf.argmax(out,1),tf.argmax(y,1))
    acc=tf.reduce_mean(tf.cast(pre_correct,'float'))
    tf.summary.scalar("acc",acc)


merged=tf.summary.merge_all()
init=tf.global_variables_initializer()
batch_size=128
# num_epo=int(mnist.train.num_examples//batch_size)

with tf.Session() as sess:
    sess.run(init)
    writer=tf.summary.FileWriter("./log",sess.graph)

    for i in range(1000):
        xx,yy=mnist.train.next_batch(batch_size)
        feed_dict={x:xx,y:yy}
        _,merge=sess.run([train_step,merged],feed_dict=feed_dict)
        writer.add_summary(merge,i)