TensorBoard 可视化 #

┌─────────────────────────────────────────────────────────────┐
│                    TensorBoard 功能                          │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  标量可视化:                                                │
│  ├── 损失曲线                                              │
│  ├── 准确率曲线                                            │
│  └── 学习率变化                                            │
│                                                             │
│  模型可视化:                                                │
│  ├── 网络结构图                                            │
│  └── 层连接关系                                            │
│                                                             │
│  直方图:                                                    │
│  ├── 权重分布                                              │
│  └── 梯度分布                                              │
│                                                             │
│  图像可视化:                                                │
│  ├── 输入图像                                              │
│  └── 特征图                                                │
│                                                             │
└─────────────────────────────────────────────────────────────┘

import keras

tensorboard = keras.callbacks.TensorBoard(
    log_dir='./logs',
    histogram_freq=1,
    write_graph=True,
    write_images=True,
    update_freq='epoch',
    profile_batch=2
)

model.fit(
    x_train, y_train,
    validation_data=(x_val, y_val),
    epochs=100,
    callbacks=[tensorboard]
)

tensorboard --logdir=./logs

keras.callbacks.TensorBoard(
    log_dir='./logs',
    histogram_freq=0,
    write_graph=True,
    write_images=False,
    write_steps_per_second=False,
    update_freq='epoch',
    profile_batch=2,
    embeddings_freq=0,
    embeddings_metadata=None
)

┌─────────────────────────────────────────────────────────────┐
│                    TensorBoard 参数                          │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  log_dir: 日志目录                                          │
│                                                             │
│  histogram_freq: 直方图记录频率                             │
│  └── 0 表示不记录                                          │
│                                                             │
│  write_graph: 是否记录模型图                                │
│                                                             │
│  write_images: 是否记录权重图像                             │
│                                                             │
│  update_freq: 更新频率                                      │
│  ├── 'batch': 每个 batch                                   │
│  ├── 'epoch': 每个 epoch                                   │
│  └── 整数: 每 N 个 batch                                   │
│                                                             │
│  profile_batch: 性能分析的 batch                            │
│                                                             │
└─────────────────────────────────────────────────────────────┘

import keras
import tensorflow as tf

log_dir = './logs'
writer = tf.summary.create_file_writer(log_dir)

class CustomTensorBoard(keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs=None):
        with writer.as_default():
            tf.summary.scalar('custom_metric', logs.get('custom_metric'), step=epoch)
            writer.flush()

model.fit(
    x_train, y_train,
    epochs=10,
    callbacks=[
        keras.callbacks.TensorBoard(log_dir=log_dir),
        CustomTensorBoard()
    ]
)

import keras
import tensorflow as tf

def log_images(epoch, logs):
    with writer.as_default():
        for i in range(5):
            img = x_test[i]
            tf.summary.image(f'test_image_{i}', img[tf.newaxis, ...], step=epoch)
        writer.flush()

writer = tf.summary.create_file_writer('./logs/images')

model.fit(
    x_train, y_train,
    epochs=10,
    callbacks=[
        keras.callbacks.LambdaCallback(on_epoch_end=log_images)
    ]
)

import keras

tensorboard = keras.callbacks.TensorBoard(
    log_dir='./logs',
    histogram_freq=1,
    write_graph=True,
    write_images=True
)

model.fit(
    x_train, y_train,
    validation_data=(x_val, y_val),
    epochs=10,
    callbacks=[tensorboard]
)

import keras
import tensorflow as tf
import datetime

(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 28, 28, 1).astype('float32') / 255.0
x_test = x_test.reshape(-1, 28, 28, 1).astype('float32') / 255.0
y_train = keras.utils.to_categorical(y_train, 10)
y_test = keras.utils.to_categorical(y_test, 10)

log_dir = './logs/' + datetime.datetime.now().strftime('%Y%m%d-%H%M%S')

model = keras.Sequential([
    keras.layers.Conv2D(32, 3, activation='relu', input_shape=(28, 28, 1)),
    keras.layers.MaxPooling2D(),
    keras.layers.Conv2D(64, 3, activation='relu'),
    keras.layers.MaxPooling2D(),
    keras.layers.Flatten(),
    keras.layers.Dense(128, activation='relu'),
    keras.layers.Dropout(0.5),
    keras.layers.Dense(10, activation='softmax')
])

model.compile(
    optimizer='adam',
    loss='categorical_crossentropy',
    metrics=['accuracy']
)

tensorboard = keras.callbacks.TensorBoard(
    log_dir=log_dir,
    histogram_freq=1,
    write_graph=True,
    write_images=True,
    update_freq='epoch'
)

history = model.fit(
    x_train, y_train,
    validation_data=(x_test, y_test),
    epochs=20,
    batch_size=128,
    callbacks=[
        tensorboard,
        keras.callbacks.EarlyStopping(patience=5, restore_best_weights=True)
    ]
)