New engine "Roc"

[THyer]

If you post or send me FEN and solutions for your test positions, I'll at least add them to the test cases list.

[peter]

That's not my point, Tom, I have lots and lots of test positions and they by far aren't all about tactical best move ones. I just want to know what to look at and which positions to test how, so just forgive me the non-programmer's question one more time:

Does higher Wobble mean less pruning, LMRs, any kind of deepening the search by costs of search-width like e.g nullmove- pruning or -verification?

Knowing that better I'd be able to select the postions better and look for tactcal best move solutions or e.g. Time to Eval as for tactical and or positional motives and test- lines. I'm always very interested in hash- learing by backward solving too,

[THyer]

Here is what I hope Wobble might accomplish.

Engines work by iterative search, repeating the same search at increasing depth until they have used the allotted time. Each of these searches is a lot like its predecessor. A risk is that one quiet move, chosen at a shallow depth, could produce decent results and "box out" a better quiet move -- each iteration would first search the ok move, and skip or reduce the better one.

By very slightly randomizing the move order, maybe we can limit this problem. The time-to-depth will likely increase, since more branches will be explored. But a very good outcome, once found, will still be explored in detail.

In short, it's an attempt to get some of the benefits of a wider search for cheaper than just increasing the branching ratio everywhere.

I think it might result in better positional play at the cost of some tactics.

[peter]

Thanks again, I think I "understand" as fas as possible as not understanding the code allows,

[Gusev]

Thanks!! FOSCEC Season 4 is coming soon, so I'd like to run Roc, provided it supports Syzygy and is stronger than Gull 3.1. I will need an executable for the AVX2/BMI2 architecture.

I am testing that now; don't know how much Elo impact it will have.

At first I will be releasing only binaries for Roc with TB support. I used Basil Falcinelli's Fathom code, but I had to mess with it to get a Windows build.

[THyer]

Roc 0.3 has been released. There are now three binaries:

x64 (without HNI)
x64_HNI
x86 (32-bit; has no tablebase support)

[Amateur]

Roc 0.3 has been released. There are now three binaries:

x64 (without HNI)
x64_HNI
x86 (32-bit; has no tablebase support)

link?
Thank you!

[MikeB]

Roc 0.3 has been released. There are now three binaries:

x64 (without HNI)
x64_HNI
x86 (32-bit; has no tablebase support)

link?
Thank you!

https://github.com/TomHyer/Roc/releases

[THyer]

Thanks, Michael.

[peter]

An UCI- option to switch off or graduate usage of tbs would be fine now too.
Sorry for being immodest