Question

当我让模型“自行构建”时，得到的结果与当呼叫Model.build()。

在下面的代码中，似乎在调用：

Model.build (<input_shape>)

没有充分利用input_shape。

在此示例中，model_input_shape和model_build不是一模一样。特别地，model_build的层不即使已传递input_shape，也知道其input_shape 到build()。

代码如下：

import tensorflow as tf

print (tf.VERSION)

model_input_shape = tf.keras.models.Sequential([
  tf.keras.layers.Dense(64, input_shape = (8, ), activation=tf.nn.relu),
])
model_input_shape.summary()
model_input_shape.weights
model_input_shape.layers[0].input_shape

model_build = tf.keras.models.Sequential([
  tf.keras.layers.Dense(64, activation=tf.nn.relu),
])
model_build.build((None, 8))
model_build.summary()
model_build.weights
model_build.layers[0].input_shape

这是运行该代码的结果：

>>> import tensorflow as tf
>>>
>>> print (tf.VERSION)
1.12.0
>>>
>>> model_input_shape = tf.keras.models.Sequential([
...   tf.keras.layers.Dense(64, input_shape = (8, ), activation=tf.nn.relu),
... ])
>>> model_input_shape.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
dense (Dense)                (None, 64)                576
=================================================================
Total params: 576
Trainable params: 576
Non-trainable params: 0
_________________________________________________________________
>>> model_input_shape.weights
[<tf.Variable 'dense/kernel:0' shape=(8, 64) dtype=float32>, <tf.Variable 'dense/bias:0' shape=(64,) dtype=float32>]
>>> model_input_shape.layers[0].input_shape
(None, 8)
>>>
>>> model_build = tf.keras.models.Sequential([
...   tf.keras.layers.Dense(64, activation=tf.nn.relu),
... ])
>>> model_build.build((None, 8))
>>> model_build.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
dense_1 (Dense)              multiple                  576
=================================================================
Total params: 576
Trainable params: 576
Non-trainable params: 0
_________________________________________________________________
>>> model_build.weights
[<tf.Variable 'dense_1/kernel:0' shape=(8, 64) dtype=float32>, <tf.Variable 'dense_1/bias:0' shape=(64,) dtype=float32>]

>>> model_build.layers[0].input_shape
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\IBM_ADMIN\Documents\wnn\python_3_7_2\lib\site-packages\tensorflow\python\keras\engine\base_layer.py", line 1338, in input_shape
    raise AttributeError('The layer has never been called '
AttributeError: The layer has never been called and thus has no defined input shape.
>>>

Model.build()实际做什么（而不是做什么）？是否有可能使用Model.build()建立模型的方式与建立模型相同 “自己建造？”

是否可以通过Model.build（）完全控制模型的构建？

0 个答案: