MSR improvements #1
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
expecting not exceed 5*std of variable
…yReLU, PReLU etc), added more tests
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Hi!
I want to collaborate in your development of MSR weight filler.
I added support for inner product layer. Current code doesn't work for ip because fan_in and fan_out are computed differently for ip and conv weight blobs.
Because ip and conv layers can have same blob dimension(100->100 ip, kernelsize = 100, in and out channels = 1, so we can't detect blob type by looking ant first 2 dimensions of blob) with different meaning, added optional FillerLayerType layer_type to FillerParameter message.
Added support for non-zero negative slope ReLU as described in original paper.
Also i added scale field to FillerParameter message.
Scale is needed in case scale of network inputs and outputs are not same ( in original paper they use scale<1 for top ip layers), or we want to manualy change amplitude of network's signal in any place.
Maybe Xavier filler also need support for this field.
Tests now expects all random variable's deviations not to exceed 5*std.