ConvLayer with_bias=False default is inconsistent to LinearLayer with_bias=True default

[albertz]

I don't really know if we should do anything now, or change this.
I just want to raise this issue here.
Maybe we can also just close it if this is fine or we cannot really do anything about it now.

I also want to know whether one of the settings is very unexpected so users might have used it wrongly. That would be an argument to make it explicit or change the default (of course via new behavior version (#508)).

Also, as usual, we should check the defaults of other frameworks.

• PyTorch Linear: bias=True
• PyTorch Conv1d: bias=True
• Keras Dense: use_bias=True
• Keras Conv1D: use_bias=True

So in all cases, it seems that with_bias=True would be more expected.

[albertz]

There was not so much feedback yet (also not on Slack).

I think with_bias=True makes more sense to have it consistent to everything else.

The question is how to introduce this. Just via a new behavior version?

[vieting]

I agree with_bias=True is more consistent. A new behavior version makes sense to me.