Prophet 4.0 and chess4j 4.0 are released

[jswaff]

After a few years of inactivity, I'm happy to simultaneously release new versions of both of my chess programs.

Prophet 4. 0 - https://github.com/jswaff/prophet

chess4j 4.0 - https://github.com/jswaff/chess4j


What's new:

Prophet has been completely rewritten. It's an entirely new codebase. (I hope to never do that again.) This has been an on-and-off-again effort over the last 2.5 years. So, in terms of algorithms, hardly anything is new. The biggest thing I can think of is changing the weights of pieces to align with Larry Kaufman's "The Evaluation of Material Imbalances" -- https://archive.is/20120630212329/http: ... alance.htm. Draw by rep code was adjusted to call "draw" after one repetition within the search. Other than that, it's pretty much the same, just cleaner and better tested. In a head-to-head match, Prophet4 is +75 ELO over Prophet3.

chess4j is now pretty much a Java port of Prophet. The algorithms are exactly the same, though the data structures are obviously different. A JNI layer was added so that Prophet can be bundled as a native engine (Linux only). In fact, this is how I usually run Prophet these days. I'm pretty sure chess4j 4.0 would be stronger than 3.5, but I haven't measured. Strength isn't really the goal of this project.

Next steps- I know everyone is jumping on the NNUE bandwagon these days (and I hope to myself before too long), but I plan to spend some time improving my evaluation, starting with some automated tuning (Texel).

[Gabor Szots]

Thanks James. I've been eagerly waiting for this version to arrive.

[CMCanavessi]

Nice! Prophet is back!

[Graham Banks]

Thanks, James.

[mvanthoor]

After a few years of inactivity, I'm happy to simultaneously release new versions of both of my chess programs.

Good

Prophet 3 has been a stable testing partner for my current version of Rustic. The thing I find strange about Prophet is that it is not affected by speed increase of Rustic. So... I test version X against engine A and B, and I get a performance rating of 2140 and 2170 respectively. Then I improve my engine to version Y (only difference being that Y is faster), which gains 20-25 Elo against both engine A and B... but not against Prophet 3. I've seen that three times in a row.

Rustic version 3.1.112 scored exactly equal in a 2000 games match against Prophet 3. Version 3.17.100 scores better against other engines than version 3.1.112, but against Prophet 3, it still scores equal.

If Prophet 4 is as stable as 3, I'll surely add it to my testing pool.

[jswaff]

After a few years of inactivity, I'm happy to simultaneously release new versions of both of my chess programs.

If Prophet 4 is as stable as 3, I'll surely add it to my testing pool.

It should be just as stable. It's played probably millions of games during the course of development without a crash.