TensorBoard
是一款为了更方便TensorFlow程序的理解、调试与优化发布的可视化工具,可以用TensorBoard来展现你的TensorFlow图像,绘制图像生成的定量指标图以及附加数据。
- 通过读取TensorFlow的事件文件来运行
- TensorFlow事件文件包括了你在运行中涉及到的主要数据
主要内容
- 通过tf.keras回调函数使用tensorboard
- 认识tensorboard界面
- 在tf.keras回调函数中记录自定义变量
- 在自定义循环中使用tensorboard
显示方式
- notebook
# 在notebook中显示
%load_ext tensorboard
%matplotlib inline
# 启动tensorboard:参数为目录地址,此处目录为logs
%tensorboard --logdir logs
- 浏览器
(1)打开cmd命令行,进入代码文件所在路径
(2)运行tensorboard --logdir logs命令
(3)将显示的网址复制到浏览器打开
记录自定义标量值
重新调整回归模型并记录自定义学习率。
- 使用创建文件编写器:tf.summary.create_file_writer()
- 定义自定义学习率功能,传递给Keras LeariningRateScheduler回调
- 在学习率功能内,用于tf.summary.scalar()记录自定义学习率
- 将LeariningRateScheduler回调传递给Model.fit()
实例:
import tensorflow as tf
import datetime
(train_images,train_labels),(test_images,test_labels) = tf.keras.datasets.mnist.load_data()
train_images = tf.expand_dims(train_images,-1)
test_images = tf.expand_dims(test_images,-1)
train_images = tf.cast(train_images/255,tf.float32)
test_images = tf.cast(test_images/255,tf.float32)
train_labels = tf.cast(train_labels,tf.int64)
test_labels = tf.cast(test_labels,tf.int64)
dataset = tf.data.Dataset.from_tensor_slices((train_images,train_labels))
test_dataset = tf.data.Dataset.from_tensor_slices((test_images,test_labels))
# 在使用fit时repeat()需要带参数,自定义循环时repeat()不需要带参数
dataset = dataset.repeat().shuffle(60000).batch(128)
test_dataset = test_dataset.repeat().batch(128)
model = tf.keras.Sequential([
tf.keras.layers.Conv2D(16,[3,3],activation='relu',input_shape=(None,None,1)),
tf.keras.layers.Conv2D(32,[3,3],activation='relu'),
tf.keras.layers.GlobalMaxPooling2D(),
tf.keras.layers.Dense(10,activation='softmax')
])
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
# 为目录名添加时间
import os
log_dir = os.path.join('logs',datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))
# 第一个参数:事件文件存放地址
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir,histogram_freq=1)
# 创建一个文件编写器
file_writer = tf.summary.create_file_writer(log_dir + '/lr')
# 设置为默认的文件编写器
file_writer.set_as_default()
# 定义方法改变学习速率
def lr_sche(epoch):
learning_rate = 0.2
if epoch > 5:
learning_rate = 0.02
if epoch > 10:
learning_rate = 0.01
if epoch > 20:
learning_rate = 0.005
#将变化写入磁盘
tf.summary.scalar('learning_rate',data = learning_rate,step = epoch)
return learning_rate
# 改变学习速率的回调函数
lr_callback = tf.keras.callbacks.LearningRateScheduler(lr_sche)
model.fit(dataset,epochs=25,
steps_per_epoch=60000//128,
validation_data=test_dataset,
validation_steps=10000//128,
callbacks=[tensorboard_callback,lr_callback]
)
在自定义训练中使用TensorBoard
optimizer = tf.keras.optimizersmizerstimizers.Adam()
loss_func = tf.keras.losses.SparseCategoricalCrossentropy()
def loss(model,x,y):
y_ = model(x)
return loss_func(y,y_)
# 初始化计算对象
train_loss = tf.keras.metrics.Mean('train_loss')
train_accuracy = tf.keras.metrics.SparseCategoricalAccuracy('train_accuracy')
test_loss = tf.keras.metrics.Mean('test_loss')
test_accuracy = tf.keras.metrics.SparseCategoricalAccuracy('test_accuracy')
def train_step(model,images,labels):
with tf.GradientTape() as t:
pred = model(images)
loss_step = loss_func(labels,pred)
#loss_step = loss(model,images,labels) # 每一步的损失值
grads = t.gradient(loss_step,model.trainable_variables) # 损失函数与可训练参数之间的梯度
optimizer.apply_gradients(zip(grads,model.trainable_variables)) # 优化函数应用梯度进行优化
# 汇总计算平均loss
train_loss(loss_step)
# 汇总计算平均acc
train_accuracy(labels,pred)
def test_step(model,images,labels):
pred = model(images)
loss_step = loss_func(labels,pred)
# 汇总计算平均loss
test_loss(loss_step)
# 汇总计算平均acc
test_accuracy(labels,pred)
current_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
train_log_dir = 'logs/gradient_tape' + current_time + '/train'
test_log_dir = 'logs/gradient_tape' + current_time + '/test'
train_writer = tf.summary.create_file_writer(train_log_dir)
test_writer = tf.summary.create_file_writer(test_log_dir)
def train():
for epoch in range(10):
for (batch,(images,labels)) in enumerate(dataset):
train_step(model,images,labels)
with train_writer.as_default():
tf.summary.scalar('loss',train_loss.result(),step = epoch) # 收集标量值,记录
tf.summary.scalar('acc',train_accuracy.result(),step = epoch)
print('Epoch{} loss is {}, accuracy is {}'.format(epoch,train_loss.result(),train_accuracy.result()))
for (batch,(images,labels)) in enumerate(test_dataset):
test_step(model,images,labels)
print('*',end='')
with test_writer.as_default():
tf.summary.scalar('loss',test_loss.result(),step = epoch) # 收集标量值,记录
tf.summary.scalar('acc',test_accuracy.result(),step = epoch)
print('Epoch{} test_loss is {}, test_accuracy is {}'.format(epoch,test_loss.result(),test_accuracy.result()))
# 重置
train_loss.reset_states()
train_accuracy.reset_states()
test_loss.reset_states()
test_accuracy.reset_states()