Skip to content

Weight Initialization, GNum class and NFData instances - #69

Open
schnecki wants to merge 206 commits into
HuwCampbell:masterfrom
schnecki:master
Open

Weight Initialization, GNum class and NFData instances#69
schnecki wants to merge 206 commits into
HuwCampbell:masterfrom
schnecki:master

Conversation

@schnecki

Copy link
Copy Markdown
Contributor

see #51

@HuwCampbell

Copy link
Copy Markdown
Owner

Sorry, I'll get round to this soon.

@schnecki

Copy link
Copy Markdown
Contributor Author

No worries, it took me quite some time as well. Btw, I checked the implemented learning algorithm and it seems to me that this is SGD with momentum and regularization. Is this correct?

Comment thread src/Grenade/Layers/Convolution.hs Outdated
Comment thread src/Grenade/Layers/Deconvolution.hs Outdated
Comment thread src/Grenade/Layers/Concat.hs Outdated
Comment thread src/Grenade/Core/Layer.hs
-- | Class for random initialization of a layer. This enables to use
-- various initialization techniques for the networks. Every layer
-- needs to implement this.
class RandomLayer x where

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does pulling this into a separate class buy us anything?

Maybe UpdateLayer should just be renamed to BaseLayer or something and hold both?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I wanted to keep UpdateLayer for backward compatibility. But renaming is fine as well. I am currently on vacation and look into the other issues soon.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants