Add SeparableConv1D layer

[zaleslaw]

Currently, the support for SeparableConv1D is missed, and it would be great to add support for this layer. The desired PR addressing this issue should include:

• Implementation of layer class named as SeparableConv1D (you can take inspiration from the implementation of Conv1D and SeparableConv2D as reference)
• Common hierarchy of all SeparableConv layers with abstract class with the common functionality
• Documentation of layer and all non-private methods
• JUnit tests in api module
• Support for export of layer to JSON (see ModelSaver.kt)
• Support for import of layer from JSON (see ModelLoader.kt)

Also, if needed, you can take a look at Keras documentation for SeparableConv1D .

NOTE: for the moment, there is no need to add support for "data format" (i.e., channels last vs. channels first) in your PR; you can assume the channels are always in the last dimension.

P.S. If you want to take this ticket, please leave the comment below
P.P.S Read the Contributing Guidelines.

[dosier]

I might be able to do this, not quite sure if I understand it well enough to do this properly.

Is SerpableConv supposed to override AbstractConv or stand on its own? I take it should override. I noted in the Keras code that SerpableConv has support for two kernels (depthwise_kernel and pointwise_kernel), so should the parent kernel in AbstractConv be ignored here?

[zaleslaw]

I suggest creating AbstractSeparableConv looking to the AbstractConv, but without inheritance.
There are a lot of different fields there: initializers, regularizers, kernels.

If it makes sense, we merge these class hierarchies.

I'll assign this ticket to you; please notify me if you will not do this.

[dosier]

I will see if I can get this one done in the weekend.

[zaleslaw]

Hi, @dosier Are you interested in continuing working on this issue?