tensorflow学习（11）- 使用tensorboard可视化手写数字识别训练过程

前言

上一节可视化了训练的网络的权值偏置值等数据的曲线图，这节可视化网络的训练过程。

代码

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.contrib.tensorboard.plugins import projector

#读取数据集
mnist = input_data.read_data_sets("../MNIST_DATA", one_hot=True)
max_steps=1001#运行次数
image_num=3000#图片数量
DIR='E:/py_code'#文件路径

sess=tf.Session()#定义会话

#载入图片（将3000个图片打包成一个矩阵）
embedding=tf.Variable(tf.stack(mnist.test.images[:image_num]),trainable=False,name='embedding')

#参数概要
def variable_summaries(var):  # 在tensorboard中显示var的相关属性值
    with tf.name_scope("summaries"):
        mean = tf.reduce_mean(var)  # 平均值
        tf.summary.scalar("mean", mean)
        with tf.name_scope("stddev"):
            stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
        tf.summary.scalar("stddev", stddev)  # 标准差
        tf.summary.scalar("max", tf.reduce_max(var))  # 最大值
        tf.summary.scalar("min", tf.reduce_min(var))  # 最小值
        tf.summary.histogram("histogram", var)  # 直方图
#命名空间
with tf.name_scope('input'):
    x=tf.placeholder(tf.float32,[None,784],name='x_input')
    y=tf.placeholder(tf.float32,[None,10],name='y_input')
#显示10个数字的图片
with tf.name_scope('input_reshape'):
    image_shaped_input=tf.reshape(x,[-1,28,28,1])
    tf.summary.image('input',image_shaped_input,10)
#创建简单神经网络
with tf.name_scope('layer'):
    with tf.name_scope('weights'):
        w=tf.Variable(tf.random_normal([784,10]),name='W')
        variable_summaries(w)
    with tf.name_scope('biases'):
        b=tf.Variable(tf.zeros([10]),'b')
        variable_summaries(b)
    with tf.name_scope('Wx_plus_b'):#线性
        Wx_plus_b=tf.matmul(x,w)+b
    with tf.name_scope('softmax'):#通过softmax激活函数变成非线性
        y_pred=tf.nn.softmax(Wx_plus_b)
#交叉熵代价函数
with tf.name_scope('loss'):
    loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y,logits=y_pred))
    tf.summary.scalar('loss',loss)
#精确度
with tf.name_scope('accuracy'):
    accuracy=tf.reduce_mean(tf.cast(tf.equal(tf.argmax(y,1),tf.argmax(y_pred,1)),tf.float32))# argmax返回一维张量中最大值所在的位置
    tf.summary.scalar('accuracy',accuracy)
#使用梯度下降算法来训练，使得loss最小（可优化）
with tf.name_scope('train'):
    train_step=tf.train.GradientDescentOptimizer(0.2).minimize(loss)

#初始化变量
sess.run(tf.global_variables_initializer())
#产生metadata文件
if tf.gfile.Exists(DIR + '/projector/projector/metadata.tsv'):
    tf.gfile.DeleteRecursively(DIR + '/projector/projector/metadata.tsv')
with open(DIR + '/projector/projector/metadata.tsv', 'w') as f:
    labels = sess.run(tf.argmax(mnist.test.labels[:], 1))
    for i in range(image_num):
        f.write(str(labels[i]) + '\n')

merged = tf.summary.merge_all()

projector_writer = tf.summary.FileWriter(DIR + '/projector/projector', sess.graph)#定义一个writer，写入图
# writer=tf.summary.FileWriter(DIR+'/projector/projector/logs',sess.graph)
saver = tf.train.Saver()#保存图
config = projector.ProjectorConfig()#配置
embed = config.embeddings.add()
embed.tensor_name = embedding.name
embed.metadata_path = DIR + '/projector/projector/metadata.tsv'
embed.sprite.image_path = DIR + '/mnist_10k_sprite.png'
embed.sprite.single_image_dim.extend([28, 28])
projector.visualize_embeddings(projector_writer, config)

#开始训练
for i in range(max_steps):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
    run_metadata = tf.RunMetadata()
    summary, _ = sess.run([merged, train_step], feed_dict={x: batch_xs, y: batch_ys}, options=run_options,
                          run_metadata=run_metadata)
    projector_writer.add_run_metadata(run_metadata, 'step%03d' % i)
    projector_writer.add_summary(summary, i)

    if i % 100 == 0:#每100个图片打印一次结果
        acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})
        print('Iter ' + str(i) + ', accuracy ' + str(acc))

saver.save(sess, DIR + '/projector/projector/a_model.ckpt', global_step=max_steps)
projector_writer.close()
sess.close()

1、首先要使用这个代码需要修改这个文件路径：

DIR='E:/py_code'#文件路径

2、然后在文件路劲下创建两级projector子目录（不然会报错找不到路径，但是网课说的是会自动创建至于为什么不行也就没有深究）：

3、修改你的图片路径：

embed.sprite.image_path = DIR + '/mnist_10k_sprite.png'

我这里就是直接放在py_code底下了，图片有1.5M。。。csdn上传图片有水印所以就放在csdn的资源里吧。
下载链接：训练图片下载传送门

可视化

1、程序运行后可以在projector子目录下看到生成的文件：

2、win + R打开命令运行窗口，输入cmd回车打开shell：

3、切换到文件对应的盘：

e:

4、运行tensorboard可视化：

tensorboard --logdir=E:\py_code\tensorflow\logs

5、复制最底下的url：http://DESKTOP-V48600Q:6006在浏览器打开，用localhost加端口号6006也能打开。
6、切换到EMBEDDINGS栏如下图：

7、选择T-SNE:

8、点击左侧中间部位的Re-run可以重新运行模型，将Color by一栏选择lable 10 colors可以区分数字的颜色：

9、训练400次的结果：

10、修改优化器提高准确度：

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.contrib.tensorboard.plugins import projector

#读取数据集
mnist = input_data.read_data_sets("../MNIST_DATA", one_hot=True)
max_steps = 1001 # 运行次数
image_num = 3000 # 图片数量
DIR = 'E:/py_code' # 文件路径

sess = tf.Session() # 定义会话

# 载入图片（将3000个图片打包成一个矩阵）
embedding=tf.Variable(tf.stack(mnist.test.images[:image_num]),trainable=False,name='embedding')

# 命名空间
x = tf.placeholder(tf.float32, [None, 784], name='x_input')
y = tf.placeholder(tf.float32, [None, 10], name='y_input')
# dropout的所要训练神经元的比例
keep_prob = tf.placeholder(tf.float32)
# 学习率
lr = tf.Variable(0.001, dtype=tf.float32)

# 显示10个数字的图片
image_shaped_input=tf.reshape(x, [-1, 28, 28, 1])
tf.summary.image('input', image_shaped_input, 10)

# 创建简单神经网络
W1 = tf.Variable(tf.truncated_normal([784, 500], stddev=0.1))
b1 = tf.Variable(tf.zeros([500]) + 0.1)
L1 = tf.nn.relu(tf.matmul(x, W1) + b1)

W2 = tf.Variable(tf.truncated_normal([500, 300], stddev=0.1))
b2 = tf.Variable(tf.zeros([300]) + 0.1)
L2 = tf.nn.relu(tf.matmul(L1, W2) + b2)

W4 = tf.Variable(tf.truncated_normal([300, 10], stddev=0.1))
b4 = tf.Variable(tf.zeros([10]) + 0.1)
y_pred = tf.nn.softmax(tf.matmul(L2, W4) + b4)


# 交叉熵代价函数
loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y,logits=y_pred))
# 精确度
accuracy=tf.reduce_mean(tf.cast(tf.equal(tf.argmax(y,1),tf.argmax(y_pred,1)),tf.float32)) # argmax返回一维张量中最大值所在的位置
# 使用梯度下降法训练，使得loss最小（#可优化）
train_step = tf.train.AdamOptimizer(lr).minimize(loss)

# 初始化变量
sess.run(tf.global_variables_initializer())
# 产生metadata文件
if tf.gfile.Exists(DIR + '/projector/projector/metadata.tsv'):
    tf.gfile.DeleteRecursively(DIR + '/projector/projector/metadata.tsv')
with open(DIR + '/projector/projector/metadata.tsv', 'w') as f:
    labels = sess.run(tf.argmax(mnist.test.labels[:], 1))
    for i in range(image_num):
        f.write(str(labels[i]) + '\n')

projector_writer = tf.summary.FileWriter(DIR + '/projector/projector', sess.graph) # 定义一个writer，写入图
# writer=tf.summary.FileWriter(DIR+'/projector/projector/logs',sess.graph)
saver = tf.train.Saver() # 保存图
config = projector.ProjectorConfig() # 配置
embed = config.embeddings.add()
embed.tensor_name = embedding.name
embed.metadata_path = DIR + '/projector/projector/metadata.tsv'
embed.sprite.image_path = DIR + '/mnist_10k_sprite.png'
embed.sprite.single_image_dim.extend([28, 28])
projector.visualize_embeddings(projector_writer, config)

# 开始训练
for i in range(max_steps):
    sess.run(tf.assign(lr, 0.001 * (0.95 ** i)))
    batch_xs, batch_ys = mnist.train.next_batch(100)
    run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
    run_metadata = tf.RunMetadata()
    summary = sess.run(train_step, feed_dict={x: batch_xs, y: batch_ys}, options=run_options,
                          run_metadata=run_metadata)
    projector_writer.add_run_metadata(run_metadata, 'step%03d' % i)

    if i % 100 == 0: # 每100个图片打印一次结果
        acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})
        print('Iter ' + str(i) + ', accuracy ' + str(acc))

saver.save(sess, DIR + '/projector/projector/a_model.ckpt', global_step=max_steps)
projector_writer.close()
sess.close()

这里注意下再次运行代码的时候需要把这些文件全删了（不然会运行报错）：