TensorBoard keras不使用训练数据显示分布和直方图

Question

我想在训练模型时监控坡度和权重。这是重现该问题的简单示例。看来keras deos并没有记录分布和直方图。 我在Ubuntu 18.04上使用Tensorflow 1.14和keras 2.4

import numpy as np
import pandas as pd

data = np.random.rand(200, 2)
expected = np.random.randint(2, size=200).reshape(-1, 1)

dataFrame = pd.DataFrame(data, columns=['a', 'b'])
expectedFrame = pd.DataFrame(expected, columns=['expected'])

dataFrameTrain, dataFrameTest = dataFrame[:100], dataFrame[-100:]
expectedFrameTrain, expectedFrameTest = expectedFrame[:100], expectedFrame[-100:]


def generator(X_data, y_data, batch_size):
    samples_per_epoch = X_data.shape[0]
    number_of_batches = samples_per_epoch / batch_size
    counter = 0

    while 1:

        X_batch = np.array(X_data[batch_size * counter:batch_size * (counter + 1)]).astype('float32')
        y_batch = np.array(y_data[batch_size * counter:batch_size * (counter + 1)]).astype('float32')
        counter += 1
        yield (X_batch, y_batch)

        # restart counter to yeild data in the next epoch as well
        if counter >= number_of_batches:
            counter = 0


from tensorflow.keras.models import Model
from tensorflow.keras.layers import Dense, Input

img_input = Input(shape=(2))

l1 = Dense(12, activation='relu')(img_input)
l2 = Dense(1, activation='sigmoid')(l1)
model = Model(inputs=[img_input], outputs=[l2])
model.compile(loss='binary_crossentropy', optimizer='adadelta', metrics=['accuracy'])

# Train the model using generator vs using the full batch
batch_size = 8
# Tensorboard callbacks
from tensorflow.keras.callbacks import TensorBoard
tb = TensorBoard(log_dir='./logs',
                 histogram_freq=1,
                 write_graph=True,
                 write_images=True,
                 update_freq='epoch',
                 write_grads=True,
                 profile_batch=1)
# train with fit_generator

model.fit_generator(generator(dataFrameTrain, expectedFrameTrain, batch_size), epochs=3,
                    steps_per_epoch=dataFrame.shape[0] / batch_size, callbacks=[tb]) 

如果我指定validation_data，它将起作用：

            validation_data=generator(dataFrameTest, expectedFrameTest, batch_size * 2),
            validation_steps=dataFrame.shape[0] / batch_size * 2

那么，为什么以这种方式实现？并且是否根据验证数据或训练数据来收集所有统计信息？！