Missing INSTALL.md about deep-pink HOT 2 CLOSED

BTW I just wanted to briefly figure out how easy/difficult it would be to add support for three new learning modes:

1. let deep pink play against itself, detecting when f(r) < f(p) and using that as a counter-example,
2. let deep pink play against sunfish, detecting the same
3. let deep pink analyze games and detect where it is obviously off

I thought that it might make sense to train on the largest violations of the f(r) > f(q) assumptions first, and maybe also try to concentrate on the last moves of the available games, to make sure that deep pink has a good chance.

I guess my Python is not good enough, though... so I won't be able to play with those ideas.

from deep-pink.

You can clone sunfish from https://github.com/thomasahle/sunfish and add it to the PYTHONPATH environment variable.

Good ideas (have been thinking about doing the same thing). Should be possible to do. The main issue is you need to generate many millions of positions for it to be able to learn efficiently, so you need to run lots of games to mine the hard negatives.

from deep-pink.