Cutsomizable "channel" dimension for ND-tensor #25
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is one of the step of generalizing 4D tensor to ND tensor for Mocha. Originally, the 3rd dimension of a 4D tensor is named the channel dimension, and many layers operate explicitly along that dimension (e.g.
SoftmaxLayer
,SoftmaxLossLayer
,AccuracyLayer
, etc.). Also,ChannelPoolingLayer
is generalized to be a 1D pooling along any user specified dimension.This PR is an effort for a well-defined behavior under the case of ND tensor. The goal is to allow the user to specify which dimension to operate on for those layers, and use a default dimension that is backward compatible with the good old 4D-tensor world.