Hi
I’m trying to understand and implement the ideas presented in W3 about customizing layers.
I’m trying to manipulate the inputs of the layer through the call and other functions within the layer class, but the initialization of the model (and by extension- the customized layer) where the batch_size is unknown, prevents from any use of this value (since it is set to “None”).
I guess there’s something basic I miss on how to handle that.
Can you offer some guidance/reference please?
We need more clarity on what you are actually asking? A good start though would be to find the code of the layer class from tensorflow (it should be on github) and try to read and understand especially the parts you are concerned with.
@muly_yahav, I’m not sure if this is what you’re running into, but one thing to be aware of with Tensorflow, is that you can use None to represent the size of a dimension that is not known in advance (like batch size).
Here’s a blog that gives an overview: Understanding Tensorflow's tensors shape: static and dynamic – P. Galeone's blog