“统一”使用什么初始化器？

Question

Keras提供各种initializers的权重和偏差。 'uniform'使用哪个？

我认为应该是RandomUniform，但这在文档中没有得到确认，并且我在源代码中陷入了死胡同：键'uniform'被用作模块和I cannot find where the variable uniform is set。

Answer 1

不过，我认为today的答案更好。

更简单的解决方案：

在交互式提示下，

import keras
keras.initializers.normal
# Out[3]: keras.initializers.RandomNormal

keras.initializers.uniform
# Out[4]: keras.initializers.RandomUniform

原始帖子：

将调试器运行到deserialize method in initializers.py 并检查

globals()['uniform']

表明该值确实是

<class 'keras.initializers.RandomUniform'>

类似地，'normal'在调试器中显示为<class 'keras.initializers.RandomNormal'>。

Note that uniform often works better than normal, and the theoretical advantages of one over the other is not clear。

Answer 2

确认这一点的另一种方法是查看initializers source code：

# Compatibility aliases

zero = zeros = Zeros
one = ones = Ones
constant = Constant
uniform = random_uniform = RandomUniform
normal = random_normal = RandomNormal
truncated_normal = TruncatedNormal
identity = Identity
orthogonal = Orthogonal