deepforge-dev/deepforge-keras

Always return attention coefficient in GraphAttention layer

Closed this issue · 2 comments

Layers that have multiple return values customized via a constructor argument (such as return_attn_coef) should always return the value and not expose the parameter to the user. The user can opt out of using the output by simply not connecting any output to it.

return_mask is another argument that should always be true.