Toga 1.4.2JD released!

[jd1]

Hi all,

I have decided to share Toga 1.4.2JD. It is in the CCRL forum under the heading Toga 1.4.2JD.

Thanks to Denis Mendoza for the compile.

All the best,
J Donald

[AdminX]

Here is a Toga142JD! link to save searching. Thanks Donald!

[Ovyron]

Thanks for sharing!

I hope to have time to test it on the future

[gerold]

Hi all,

I have decided to share Toga 1.4.2JD. It is in the CCRL forum under the heading Toga 1.4.2JD.

Thanks to Denis Mendoza for the compile.

All the best,
J Donald

Thanks J.D. I will be testing it vs.Strelka2 today.. Should be interesting.

Best to you,

Gerold.

[Denis P. Mendoza]

Hi all,

I have decided to share Toga 1.4.2JD. It is in the CCRL forum under the heading Toga 1.4.2JD.

Thanks to Denis Mendoza for the compile.

All the best,
J Donald

It's nice that you made it publicly available. And you're welcome too.

[swami]

Thanks for updating it. Any idea about it's rating?

[Leto]

Too early to say but my initial impression is that the 4CPU version is weaker in blitz than the 4CPU version of Toga II 1.4.1SE.

Against Naum 3 x64 4CPU it scored only 31% in 50 games, compared to 46% scored by 1.4.1SE. Against Zappa Mexico x64 4CPU it is scoring 31% after 20 games, compared to 39% scored by 1.4.1SE.

Again it is way too early to make any accurate determination on the engine's strength.

[Leto]

A quick update on my 1.4.2JD vs Zappa match, after 42 games 1.4.2JD improved its score to 39.29%.

[mjlef]

Hi all,

I have decided to share Toga 1.4.2JD. It is in the CCRL forum under the heading Toga 1.4.2JD.

Thanks to Denis Mendoza for the compile.

All the best,
J Donald

Thanks! Some very interesting change. I looked over the code and noticed some changes from earlier Togas in Late Move Reduction:

a. it no longer uses the fail high statistics in making the reduction decision.
b. It reduces in the PV after 6 moves (Glaurung and some other do something similar)
c. It reduced sacrificial moves more (which seems to be defined as a move to a quare which is attached, undefended and SEE determines the move is a loss, although I suppose any piec attacked and undefended is a loss anyway unless if it is a capture of a bigger piece by a smaller one)

In the case where say a non capture is doubly reduce, but ends up failing high, the research only increases the depth by one, so this move is still reduced in the research, but one ply less. Have these changes been beneficial? I only ask since I have tried similar things but have never been able to prove they help. maybe you have tested more than me.

I also see a Strelka-like form of futility where at depth<=3 and with a low score (score<alpha-300) you try a qsearch to see if something is good enough to get >=alpha. Have you shown benefits in this? I was just wondering what data you might have. I do find some modest benefits to doing this, depending on how the qsearch is implemented.

Mark

[Jouni]

I run some testsuites already. And in ALL "JD" was weaker than 1.4 - definitely noot good sign at all.

Jouni