Skip to content

Conversation

@hans-ekbrand
Copy link

Cap the probability of inclusion into training for a position, regardless of its q.diff.

This idea was used to create the net posted here: https://discord.com/channels/425419482568196106/539960268982059008/874663457830633542

@hans-ekbrand hans-ekbrand marked this pull request as draft August 10, 2021 19:16
@hans-ekbrand hans-ekbrand marked this pull request as draft August 10, 2021 19:16
@hans-ekbrand hans-ekbrand marked this pull request as draft August 10, 2021 19:16
@hans-ekbrand
Copy link
Author

The code is working, but I haven't tested the performance properly yet.

@Tilps
Copy link
Contributor

Tilps commented Aug 10, 2021

I think mathematically - this should be equivalent to increasing the slope and the min. Specifically a 0.3 max is equivalent to multiplying slope and min by 3.3. (To get very exactly equivalent also scale SKIP by 3.3 - but scaling SKIP probably isn't needed to be 'good enough equivalence'.)

@hans-ekbrand
Copy link
Author

Good point, but it is still convenient to have this parameter, especially since SKIP is not a yaml-parameter.

@hans-ekbrand hans-ekbrand marked this pull request as ready for review August 11, 2021 10:10
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.

2 participants