Generate more training data

[SongRb]

I'm interested in how to generate more training data from existing .sgf file, especially for the policy_target and value_target attribute. Is there some code already implemented this feature?

[Narsil]

What do you mean more training data ? Do you mean to have more information on what was the policy target and value_target on each move ?

You can add them in the comments as there are already some information in there. (The full policy is a bit hard to read which is why I removed it). Also, I am looking into adding this info into sabaki (via the gtp play) to have prettier information while it is playing. https://github.com/yishn/LeelaSabaki

Note, that I won't implement that right away as it seems that my network is a bit stuck, so I will probably look into why first.

Btw if you are using sgf files, I do recommend sabaki, it is the less annoying one concerning self sucide (which is a legal move with the Tromp-Taylor ruleset)