Honey-XPro Release

[ChiefPushesWood]

Would it be possible for someone to explain exactly what tactical does in the settings? Or point me to documentation?

CPW

[Ovyron]

Would it be possible for someone to explain exactly what tactical does in the settings? Or point me to documentation?

Tactical uses MultiPV internally, as a power of two. So Tactical=1 is MultiPV=2. Tactical=2 is MultiPV=4. Tactical=3 is MultiPV=8, and so on.

There's no difference between using Tactical and using MultiPV with those values.

Except Tactical hides the scores of the other moves it analyzes, so it only shows the score of the best move. It's not useful for interactive analysis (where the score of those moves is important), it's only useful for when Honey would not find the best move in Single PV, and requires Tactical to find it (but MultiPV of equivalent value would do the same and show scores of other moves.)

[MBrain's, McCain's (Tactical setting) and Shash's Deep Analysis Mode work in the same way]

[jp]

Tactical uses MultiPV internally, as a power of two. So Tactical=1 is MultiPV=2. Tactical=2 is MultiPV=4. Tactical=3 is MultiPV=8, and so on.

[MBrain's, McCain's (Tactical setting) and Shash's Deep Analysis Mode work in the same way]

There's no advantage to powers of two, is there?

[ChiefPushesWood]

Would it be possible for someone to explain exactly what tactical does in the settings? Or point me to documentation?

Tactical uses MultiPV internally, as a power of two. So Tactical=1 is MultiPV=2. Tactical=2 is MultiPV=4. Tactical=3 is MultiPV=8, and so on.

There's no difference between using Tactical and using MultiPV with those values.

Except Tactical hides the scores of the other moves it analyzes, so it only shows the score of the best move. It's not useful for interactive analysis (where the score of those moves is important), it's only useful for when Honey would not find the best move in Single PV, and requires Tactical to find it (but MultiPV of equivalent value would do the same and show scores of other moves.)

[MBrain's, McCain's (Tactical setting) and Shash's Deep Analysis Mode work in the same way]

Thank you!

[Ovyron]

There's no advantage to powers of two, is there?

The advantage is saving time when switching multiple lines (normally for MultiPV you go 1->2->3->4->5->6->7->8, with Tactical it's just 1->2->4->8.)

The disadvantage is being unable to use intermediate levels (Tactical can't use MultiPV=6.) But to me hiding the score of the other moves (when the engine is using valuable time to give them exact scores) is the deal breaker.

[Master Om]

Should we use MCTS or not? Will be the AB search better or the MCTS one ?
I found Komodo's MCTS is good for nothing.

[MikeB]

Would it be possible for someone to explain exactly what tactical does in the settings? Or point me to documentation?

Tactical uses MultiPV internally, as a power of two. So Tactical=1 is MultiPV=2. Tactical=2 is MultiPV=4. Tactical=3 is MultiPV=8, and so on.

There's no difference between using Tactical and using MultiPV with those values.

Except Tactical hides the scores of the other moves it analyzes, so it only shows the score of the best move. It's not useful for interactive analysis (where the score of those moves is important), it's only useful for when Honey would not find the best move in Single PV, and requires Tactical to find it (but MultiPV of equivalent value would do the same and show scores of other moves.)

[MBrain's, McCain's (Tactical setting) and Shash's Deep Analysis Mode work in the same way]

An undocumented feature that you can also combine tactical and multipv - say you want a wider search tactical to 3 ( search as if multipv was set to 8) and you set multipv to see just the best 4 best moves , although it is doing wider searches if was multipv was searched to 8, it will just display the 4 best moves.

[ChiefPushesWood]

@MikeB I sent you a PM..

CPW

[ChiefPushesWood]

I've looked all over Github (I think) and I can't find a detail explanation of the various UCI settings and what they do in Honey. Is that available? (Sorry, I'm new to this stuff - that is, tweaking and understanding the UCI settings)

CPW

PS. If it matters, I'm currently using the Stanback zip for Honey. I don't know how to compile my own.

[ChiefPushesWood]

Hey there. Okay... I've been playing around with Honey today and I'm a little confused and need some help/guidance.
I have a position (I'm not going to post it here because it's an ongoing game - even though my opponent could comfortably resign but... such is life in ICCF) that I've taken a look at this morning in Chessbase.

Same position, clear hash each time and allow it to get to depth=40. Here are the settings and results:

Code: Select all

SCENARIO 1: Stockfish Latest Dev (05/24/19)
22 CPU's, 16GB Hash, Contempt 24, Analysis Contempt - White, Using 'NEXT BEST' feature in Chessbase
----------------------------
34.Qh4	+6.90	D=40
34.Be2	+6.40	D=40
34.Bb5	+6.16	D=40
34.Bf1	+6.25	D=40

SCENARIO 2: Honey XPro
22 CPU's, 16GB Hash, White Contempt 22(Default), Black Contempt 2(Default), Analysis 3, MPV set to 4 in GUI 
-----------------------------
34.Qh4	+4.95	D=40
34.Be2	+4.78	D=40
34.Bf1	+4.62	D=40
34.Bb5	+4.55	D=40

SCENARIO 3: Honey XPro (trying to match the SF settings above)
22 CPU's, 16GB Hash, White Contempt 24(to match SF), Black Contempt 0, Analysis 0, SinglePV in GUI, Using 'NEXT BEST' feature in Chessbase
-----------------------------
34.Qh4 	+4.43	D=40
34.Be2	+4.56	D=40
34.Bf1	+4.56	D=40
34.Bb5	+4.56	D=40

Alright, so after doing this little exercise, it's clear I do not understand some things and I have some questions:

1. Is the eval function different in Honey than SF? Obviously, comparing the evals of scenario 1 and 2, it's a huge difference. I thought possibly it was the MPV search of Honey holding down the eval (maybe not seeing as deep because of Tactical=3 and GUI MPV=4). So I ran the same search as scenario 1 in scenario 3 and things got even more wonky.

2. Am I misunderstanding contempt? In SF, as I understand it, with the default setting of Contempt=24, I set the Analysis contempt to WHITE because I wanted it ONLY for white (conversely, if I wanted it only for black, I'd set Analysis Contempt to BLACK). Is that correct? Now, is this different in Honey? In scenario 3, I set White Contempt to 24 and Black Contempt to 0, thinking this would be the same as SF. But, as you can see, the evals were way different and 34.Qh4 came in as the 4th best move. So I think I'm not understanding something here.

3. If I want contempt ONLY for white in Honey, do I just leave the default settings? (White Contempt=22; Black Contempt=2) - well I guess 2 gives a 'little' contempt to black. And if I want contempt for black's side, should I change those around (White Contempt=2; Black Contempt=22)?

4. Is searching a position this way (MPV and possibly using Tactical=x) really better or even as good as 'NEXT BEST'?? I understood that, in the past (years ago?), MPV on AB engines weakened analysis. Is that not the case anymore today? While SF (Scenario 1) and Honey in MPV (Scenario 2) did indeed come to the same 4 top moves, the time required to do so was quite different (as was the results). I was able to get depth 40 on all 4 lines (one at a time, using NEXT BEST) in SF and Honey (Scenario 1 and 3) in about 25 minutes on my machine. Getting to depth 40 in Scenario 2 (MPV and Tactical) took just under an hour. I certainly don't mind the time IF there is a benefit to doing it this way. So, help a fella out, is one actually better, more thorough to the other? Is there a benefit to Honey's tactical+mpv search vs. singlePV and 'NEXT BEST'?

That's all I can think of right now for this test session. I appreciate any help I can get making these things more clear and understanding what I'm seeing from this engine.

Thanks,
CPW