Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Spend less time if we have found a TB win #3186

Closed

Conversation

SFisGOD
Copy link
Contributor

@SFisGOD SFisGOD commented Oct 17, 2020

There is no need to spend lots of time if we have found a TB win so we reduce the allowable total time to 20% of optimum time.

Bench: 3834252

@SFisGOD SFisGOD changed the title Spend less time if we have found a known win Spend less time if we have found a TB win Oct 17, 2020
@syzygy1
Copy link
Contributor

syzygy1 commented Oct 18, 2020

Does this cover both the case that the root position is a TB win and the case that a TB win is found in the search? (Probably it does... I did not try to verify.)

How about doing the same or similar in case of a TB loss? (Of course only if the root position is a TB loss or if a full ply was searched so there is no move that can avoid the TB loss.)

Perhaps same if the root position is a TB draw?

@vondele
Copy link
Member

vondele commented Oct 18, 2020

CI is failing on this one.

However, some comments for consideration:
This seem to trigger in the case of having a TB position in the root. These are not the cases relevant for TCEC, I think, since games are adjudicated at that point?

Doing the more general case, i.e. once search has found a TB win would cover these cases, one would wonder if finding a mate score would be useful as well.

I think we prove winning mate scores (and probably TB wins), but I'm not so sure about losses.

If search finds a mate score or a TB win, and the position is not searched well enough, it will miss shortest wins, and in a few rare cases, will actually not be able to find it back in the next move.

So, not so sure that this patch is a good idea.

What does seem to happen at TCEC is that search takes something a very long time once a TB score has been found, as if the iteration takes forever to complete. This would need to be understood IMO. Maybe these things are related to #3170 or #2717

@SFisGOD
Copy link
Contributor Author

SFisGOD commented Oct 18, 2020

@vondele Updated. I increased it to 20% for concerns about the rare cases of losing mate score or TB win.

@vondele
Copy link
Member

vondele commented Oct 22, 2020

one side effect will be that once we have found a TB win, it will become much more rare to see a mate score. Personally, I like to see a #81 or similar to appear in a tournament, it is always mind-boggling to see how an engine can find lines that deep.

@vondele
Copy link
Member

vondele commented Nov 1, 2020

I'll not merge this one, I think searching for a better mate etc is worthwhile.

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.

None yet

3 participants