Minic version 2

[xr_a_y]

Only few games made by me...
30 sec for 1 move...
At the moment Minic 1.54 seems better...

To see anything on such close versions, you will indeed need to play at least 3000 games, which probably won't be possible at 30sec per move.
Maybe using 2.07 you can see things clearer.

Regards

[Gabor Szots]

To see anything on such close versions, you will indeed need to play at least 3000 games, which probably won't be possible at 30sec per move.
Maybe using 2.07 you can see things clearer.

I have started a gauntlet with version 2.07 under the Shredder 13 GUI. It took some time to make the engine run and use tablebases, because of this:

1. During installation there is no way to specify command line arguments, so syzygy TB's are not used.
2. When I edited the .eng file manually, putting 'Parameter=-syzygyPath ...' in it the engine did not even start, Shredder told me it was not an UCI engine.
3. Then it dawned on me that perhaps -uci should also be included in the Parameter line. And voilá, it worked.

So it seems that either you don't specify anything or you have to specify both things. I guess Shredder issues the uci command by default but the Parameter line takes preference over that.
I'd like to ask you Vivien to make syzygyPath an option, settable together with the rest of the options.

Regards,
Gabor

[xr_a_y]

To see anything on such close versions, you will indeed need to play at least 3000 games, which probably won't be possible at 30sec per move.
Maybe using 2.07 you can see things clearer.

I have started a gauntlet with version 2.07 under the Shredder 13 GUI. It took some time to make the engine run and use tablebases, because of this:

1. During installation there is no way to specify command line arguments, so syzygy TB's are not used.
2. When I edited the .eng file manually, putting 'Parameter=-syzygyPath ...' in it the engine did not even start, Shredder told me it was not an UCI engine.
3. Then it dawned on me that perhaps -uci should also be included in the Parameter line. And voilá, it worked.

So it seems that either you don't specify anything or you have to specify both things. I guess Shredder issues the uci command by default but the Parameter line takes preference over that.
I'd like to ask you Vivien to make syzygyPath an option, settable together with the rest of the options.

Regards,
Gabor

Yes, sorry for this blurry configuration of Minic.
I don't know Shredder gui but please just try this

Code: Select all

minic -uci -syzygyPath /your/path/to/TB

so that's what you do now, but you specify also that protocol is UCI.

I'll add syzygy options to classic UCI stuff soon.

[Gabor Szots]

minic -uci -syzygyPath /your/path/to/TB

Yes, that's what I did when I said 'voilá'.

[xr_a_y]

minic -uci -syzygyPath /your/path/to/TB

Yes, that's what I did when I said 'voilá'.

The patch is done, will publish 2.08 soon with that included.

[Gabor Szots]

minic -uci -syzygyPath /your/path/to/TB

Yes, that's what I did when I said 'voilá'.

The patch is done, will publish 2.08 soon with that included.

Thanks Vivien. If you haven't changed anything else then I may safely replace 2.07 with 2.08 in my ongoing tournament. Otherwise I stick with 2.07.

[xr_a_y]

minic -uci -syzygyPath /your/path/to/TB

Yes, that's what I did when I said 'voilá'.

The patch is done, will publish 2.08 soon with that included.

Thanks Vivien. If you haven't changed anything else then I may safely replace 2.07 with 2.08 in my ongoing tournament. Otherwise I stick with 2.07.

You will have to stick to 2.07 because 2.08 will introduce functional change.

[xr_a_y]

Minic 2.11 is released, taking into account recent various requests :
- mateFinder in GUI option
- syzygyPath in GUI option
- contempt in GUI option (using Alayan formulation in Ethereal, i.e. { contemptMG + contempt, contempt })
- +15 at blitz TC versus 2.07

For testing purpose, as I release many versions very fast, I suggest that CCRL40/15 go on with 2.00 testing that's seems quite good.

From there, I will try to tune my recent tricks :
- pruning / reduction depending on danger
- various margins that can be different if static score is from TT eval or just static eval
- better mixing of various history scores for move ordering
- i'd love to both get ride of spurious delaying checks and follow better good checks sequence

[xr_a_y]

I am currently trying to tune the new Minic skill feature.
Instead of my former dummy level thing, I use the Stockfish idea.
Here what I get for now based on CCRL Elo:

Minic level 0 : random mover, 280
Minic level 1 : weakest, 570
Minic level 10 : 850
Minic level 20 : 1100
Minic level 30 : 1500
Minic level 40 : 2000
Minic level 50 : 2300
Minic level 60 : 2500
Minic level 70 : 2700
Minic level 80 : 2850
Minic level 90 : 2950
Minic level 100 : 3100

I am now trying get more levels in the weak part of the Elo spectrum. I'm only playing with depth and multipv, I'll try to disable some knowledge also.
I will try to tune this so that level 60 is around 2000.

I'm also trying so fit UCI_Elo somehow, for now I use this linear thing (which is wrong ...):

level = max(1,(UCI_Elo -500)/29)

Still work on progress I guess ...

[xr_a_y]

Minic 2.16 is available and should be around 3050 single core at blitz TC (CCRL scale).
This version will be soon tested on TCEC hardware, hopefully fixing the previous known issues.
Last TCEC tested version was 1.48. There is something like 80 Elo between this two versions at very fast TC, let's see what it gives on crazy hardware at longer TC...