Closed ghost closed 8 years ago
@seyounglee95 I'll take a look. However, test build failed since NullPointException
is thrown during LayerParameterArrayCodecTest
. Please fix this :)
@seyounglee95 This looks good. I finished my pass and left some comments. Please check them :)
@beomyeol Thank you for your review. I fixed referring to your comments. Please take a look.
@seyounglee95 I finished another pass and left some comments.
@beomyeol Thank you for your review. I made some modifications. Please take a look.
@beomyeol I pushed a commit referring to your comment. Please check it.
@beomyeol I fixed comments. Thank you :D
@beomyeol Thank you for comments. I fixed declaration and comments.
@seyounglee95 Looks good. I finished my pass. @jsjason, can you take a look?
@beomyeol @seyounglee95 Yes, I'll do a pass. I'll merge this right away (given that I don't see any problems), since this PR has been open for a long time.
@seyounglee95 Great. I'll merge this after you're done checking my comments.
Thank you for review. Please take a look.
Great. I'll merge this.
As a first step for implementing CNN, this PR introduce new layer class
PoolingLayer
. This layer extendsLayerBase
, and include implementation forisLearnable()
,getOutputShape()
andgenerateParameterGradient()
. This PR include some modifications inLayerParameter
, which enablepoolingLayer
to make use of its new parameters. Also,PoolingLayerConfigurationBuilder
class andPoolingLayerParameterInitializer
class are added for configuration of pooling layer.neural_network.proto
is modified to take inputs for pooling layer. Concrete implementation forfeedForward()
andbackPropagate()
will be made in later PRs.