如何制作单个向量/数组？

Question

考虑以下cnn model

def create_model():
  x_1=tf.Variable(24)
  bias_initializer = tf.keras.initializers.HeNormal()
  model = Sequential()
  model.add(Conv2D(32, (5, 5),  input_shape=(28,28,1),activation="relu", name='conv2d_1', use_bias=True,bias_initializer=bias_initializer))
  model.add(MaxPooling2D(pool_size=(2, 2)))
  model.add(Conv2D(64, (5, 5), activation="relu",name='conv2d_2',  use_bias=True,bias_initializer=bias_initializer))
  model.add(MaxPooling2D(pool_size=(2, 2)))
  model.add(Flatten())
  model.add(Dense(320, name='dense_1',activation="relu", use_bias=True,bias_initializer=bias_initializer),)
  model.add(Dense(10, name='dense_2', activation="softmax", use_bias=True,bias_initializer=bias_initializer),)
  return model

我创建了上述 model 的model_1=create_model()实例。 现在考虑以下

combine_weights=[]
for layer in model.layers:
  if 'conv' in layer.name or 'fc' in layer.name:
    print(layer)
    we=layer.weights
    combine_weights.append(we)

从model_1 ，上面的代码获取卷积层/fc 层的权重，并将它们组合在一个combine_weight数组中。 combine_weight的 dtype 是通过print(type(combine_weights))获得类型<class 'list'>

现在，我尝试使用以下combine_weights_reshape=tf.reshape(tf.stack(combine_weights,[-1]))重塑所有这些权重以产生单行向量/一维数组，这会产生以下错误

<ipython-input-80-dee21fe38c89> in <module>
----> 1 combine_weights_reshape=tf.reshape(tf.stack(combine_weights,[-1]))

1 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py in raise_from_not_ok_status(e, name)
   7184 def raise_from_not_ok_status(e, name):
   7185   e.message += (" name: " + name if name is not None else "")
-> 7186   raise core._status_to_exception(e) from None  # pylint: disable=protected-access
   7187 
   7188 

InvalidArgumentError: Shapes of all inputs must match: values[0].shape = [5,5,1,32] != values[1].shape = [32] [Op:Pack] name: stack

如何将combine_weight重塑为单行向量/数组？

Answer 1

我通过以下方式得到了预期的结果

combine_weights=[]
con=[]
for layer in model.layers:
  if 'conv' in layer.name or 'fc' in layer.name:
    print(layer.name)
    we=layer.weights[0]
    we_reshape=tf.reshape(we,[-1])
    # bi=layer.weights[1]
    combine_weights.append(we_reshape)
    print(combine_weights)
    print(len(combine_weights))
    con=tf.concat([con,we_reshape], axis=[0])
    print(con)

Answer 2

一种解决方案是在将权重张量附加到权重列表之前对其进行展平。 最初的问题是权重张量具有不同的形状，因此tf.stack不起作用。

combine_weights = []
for layer in model.layers:
    if "conv" in layer.name or "fc" in layer.name:
        print(layer)
        # Flatten the weights tensor.
        we = tf.reshape(layer.weights, shape=-1)
        combine_weights.append(we)
# Concatenate all of the (flat) weight vectors.
combine_weights = tf.concat(combine_weights, axis=0)