Bug hunting

[jhellis3]

For memory problems is good also valgrind.

2nd this.

[Patrice Duhamel]

It's possible the inconsistency is caused by a bug but it could be something benign like the above as well. If you're trying to figure out the specific cause you should try to control for as many of these features in your experiments as you can.

I will look more closely to the games, and try to understand what happens.

If it's not a bug and the problem is only the evaluation function, maybe I should disable all the pruning + reductions when I try to tune the evaluation ?

[Ronald]

You could also work the other way around, i.e start with a minimalistic evaluation function and try to optimize/debug you search first. You could implement the PeSTO or any other PST evaluation so you can determine the base level with that (you might need to tweak some search parameters to match the PST values). rofChade 1.0 was a PST only version and the search parameters are still nearly the same (I believe only LMR is more aggressive). I guess rofChade's full evaluation function added around 300 elo to the PST only version, you might use that to compare it with the difference of your own evaluation function to get more feeling for where you might win the most elo, search or eval

[Patrice Duhamel]

You could also work the other way around, i.e start with a minimalistic evaluation function and try to optimize/debug you search first. You could implement the PeSTO or any other PST evaluation so you can determine the base level with that (you might need to tweak some search parameters to match the PST values). rofChade 1.0 was a PST only version and the search parameters are still nearly the same (I believe only LMR is more aggressive). I guess rofChade's full evaluation function added around 300 elo to the PST only version, you might use that to compare it with the difference of your own evaluation function to get more feeling for where you might win the most elo, search or eval

Yes, it's an interesting idea.