TalkChess.com

With captures, this doesn't work because they are too specific since a lot depends on whether and how things are protected. That is also what I thought. But then I don't understand how engines like Ethereal and Stockfish use capture histories? It seems - at least in Ethereal's example (https://gith...

Hi. Apart from ordering captures with SEE and MVV/LVA, I have seen some of the strongest engines utilize some kind of history heuristic just for captures. This makes sense to me considering that some positions have very thematic captures. Take for example the one below: rnbq1rk1/pppn1ppp/4p3/3pP3/1b...

New, big to-do point: As I have stated before, I would like to see all heavy inspiration and copying of Vice being gone, and to make sure I get it done, I am planning on not releasing Loki 4.0 before. The main point of inspiration from Vice is the UCI-implementation which will have to be completely...

Good efforts. Good luck with the implementation. It's hard to test this accurately, and it's hard to know what each function would give you with regard to Elo gain. Thank you :D . The reason, I am testing each feature/function this way is not to get an accurate number (since it's effect may vary gr...

Alright. Thank you very much for your help!
Search tuning has been on my mind for quite some time, so I am happy to have gotten a place to start

I just experimented with late move pruning, but unfortunately I didn't see any elo gain :( . But luckily no loss either. My implementation is the following: // Step 13. Late move pruning (~19 elo). If we are near the horizon and the highest ordered moves haven't caused a beta cutoff, we're probably ...

Alright, after testing some different ideas (especially for initial reduction values, which is still just one ply), my current LMR implementation inside PVS is: // Step 13. Principal variation search new_depth = depth - 1 + extensions; if (moves_searched == 0) { // Always search the first move at fu...

While not working on the staged move generation, I have decided to create a new branch for LMR and LMP. My goal is to gain 100-200 elo points with these features and then merge the branch into master again. Yesterday, I started with a very simple implementation from (https://web.archive.org/web/2015...

You can try the following. Chess Tuning Tools Lakas Optuna Game Parameter Tuner I developed the last 2. They can optimize search or eval params or combination thereof. Some optimizers have genetics algo both in Lakas and Optuna. Lakas uses the nevergrad framework while optuna uses the Optuna framew...

I wrote a simple NN library, if anyone is interested in developping their own trainer. https://github.com/lucasart/nn Simple code, documented, tested, easily extensible. Happy coding! What it does: Create network (fully connected MLP) with arbitrary dimensions and activation functions (linear, relu...