New engine: Stash

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Gabor Szots
Posts: 491
Joined: Sat Jul 21, 2018 5:43 am
Location: Szentendre, Hungary
Full name: Gabor Szots

Re: New engine: Stash

Post by Gabor Szots » Tue Mar 03, 2020 10:26 am

mhouppin wrote:
Tue Mar 03, 2020 9:28 am
I'm currently doing assets with compiled binaries for principal 64-bit architectures so that it becomes possible to use them without a compiler.
I have successfuly compiled Stash 10.0 for my architecture (Haswell). But that is rather the exception than the rule so original compiles by the author are always welcomed.
Gabor Szots
CCRL testing group

mhouppin
Posts: 52
Joined: Wed Feb 12, 2020 4:00 pm
Full name: Morgan Houppin

Re: New engine: Stash

Post by mhouppin » Tue Mar 03, 2020 5:01 pm

Stash Ratings List updated (110 additional matches played, time control 10m+10s/move):

Code: Select all

   # PLAYER       :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Stash 9.0    :     444    145    34.0      40  85.0      93   31    6    3  15.0
   2 Stash 8.0    :     334    136    30.0      40  75.0      65   23   14    3  35.0
   3 Stash 8.1    :     310    132    29.0      40  72.5      87   21   16    3  40.0
   4 Stash 7.0    :     242    126    26.0      40  65.0      76   17   18    5  45.0
   5 Stash 8.2    :     199    120    24.0      40  60.0      99   16   16    8  40.0
   6 Stash 6.0    :      64    117    17.5      40  43.8      81    9   17   14  42.5
   7 Stash 2.0    :      11    125    15.0      40  37.5      57    7   16   17  40.0
   8 Stash 3.0    :       0   ----    14.5      40  36.2      64    8   13   19  32.5
   9 Stash 5.0    :     -22    122    13.5      40  33.8      50    7   13   20  32.5
  10 Stash 4.0    :     -22    120    13.5      40  33.8     100    8   11   21  27.5
  11 Stash 1.0    :    -341    200     3.0      40   7.5     ---    0    6   34  15.0
Currently testing Stash 10.0 to add it to the list.

Alayan
Posts: 309
Joined: Tue Nov 19, 2019 7:48 pm
Full name: Alayan Feh

Re: New engine: Stash

Post by Alayan » Tue Mar 03, 2020 5:20 pm

More games at a shorter TC would be more meaningful. It takes a lot of computing time to generate this small amount of games with this TC.

60s+0.6s is already more than enough to compare most engines, and 5+0.05s would be enough for SPRT tests for a very weak engine.

mhouppin
Posts: 52
Joined: Wed Feb 12, 2020 4:00 pm
Full name: Morgan Houppin

Re: New engine: Stash

Post by mhouppin » Wed Mar 04, 2020 4:39 pm

Alayan wrote:
Tue Mar 03, 2020 5:20 pm
More games at a shorter TC would be more meaningful. It takes a lot of computing time to generate this small amount of games with this TC.

60s+0.6s is already more than enough to compare most engines, and 5+0.05s would be enough for SPRT tests for a very weak engine.
As you suggested, I ran a Round-Robin with a time control of 60s+0.6s, here are the current results:

Code: Select all

   # PLAYER        :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Stash 10.0    :     988    201    98.0     100  98.0     100   96    4    0   4.0
   2 Stash 9.0     :     547    106    78.5     100  78.5     100   72   13   15  13.0
   3 Stash 8.1     :     381     98    66.0     100  66.0      54   53   26   21  26.0
   4 Stash 8.0     :     377     97    65.0     100  65.0      75   54   22   24  22.0
   5 Stash 8.2     :     346     97    62.0     100  62.0      62   51   22   27  22.0
   6 Stash 7.0     :     333     94    61.0     100  61.0     100   44   34   22  34.0
   7 Stash 5.0     :     153     89    41.0     100  41.0      87   27   28   45  28.0
   8 Stash 6.0     :     103     90    35.5     100  35.5      89   22   27   51  27.0
   9 Stash 4.0     :      50     88    29.0     100  29.0      87   16   26   58  26.0
  10 Stash 3.0     :       0   ----    25.0     100  25.0      80   10   30   60  30.0
  11 Stash 2.0     :     -40     92    20.5     100  20.5      64    8   25   67  25.0
  12 Stash 1.0     :     -57     94    18.5     100  18.5     ---    9   19   72  19.0

mhouppin
Posts: 52
Joined: Wed Feb 12, 2020 4:00 pm
Full name: Morgan Houppin

Re: New engine: Stash

Post by mhouppin » Thu Mar 05, 2020 8:07 pm

Round-robin ended, here are the results:

Code: Select all

   # PLAYER        :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Stash 10.0    :     940    116   213.5     220  97.0     100  208   11    1   5.0
   2 Stash 9.0     :     581     68   176.0     220  80.0     100  161   30   29  13.6
   3 Stash 8.1     :     408     65   143.5     220  65.2      73  112   63   45  28.6
   4 Stash 8.0     :     390     65   139.5     220  63.4      77  110   59   51  26.8
   5 Stash 7.0     :     368     62   134.5     220  61.1      76   97   75   48  34.1
   6 Stash 8.2     :     348     61   130.0     220  59.1     100  101   58   61  26.4
   7 Stash 5.0     :     185     58    91.0     220  41.4      94   63   56  101  25.5
   8 Stash 6.0     :     139     59    80.0     220  36.4      89   46   68  106  30.9
   9 Stash 4.0     :     103     59    71.5     220  32.5     100   41   61  118  27.7
  10 Stash 3.0     :       0   ----    49.0     220  22.3      56   18   62  140  28.2
  11 Stash 2.0     :      -5     61    48.0     220  21.8      75   23   50  147  22.7
  12 Stash 1.0     :     -28     63    43.5     220  19.8     ---   20   47  153  21.4
I'm currently relaunching a Round-robin to reduce error bars.

mhouppin
Posts: 52
Joined: Wed Feb 12, 2020 4:00 pm
Full name: Morgan Houppin

Re: New engine: Stash

Post by mhouppin » Thu Mar 12, 2020 12:53 pm

Hi there, version 11 is out!

I simply added a Transposition Table, storing the position key, the search depth, the search evaluation, the search bound (exact result, under alpha, over beta?), the generation id, and the best move (none if the search failed under alpha). For now I'm not trusting passed generation evaluations (I got some problems with 3-fold repetition detection), only best moves found.

The analysis depth raised from 1-2 plies in the middle game, and from 2-3 plies in endgame (even going go +6 plies in positions with less than 6 pieces)!
I expect a gain of around 100 Elo with this component, and a CCRL Blitz Rating of around 1900 Elo.

I'm currently working on adding a Null Move Search (testing computations for the depth reduction and the minimal plies from null).

Gabor Szots
Posts: 491
Joined: Sat Jul 21, 2018 5:43 am
Location: Szentendre, Hungary
Full name: Gabor Szots

Re: New engine: Stash

Post by Gabor Szots » Thu Mar 12, 2020 1:39 pm

mhouppin wrote:
Thu Mar 12, 2020 12:53 pm
Hi there, version 11 is out!
Is it not supposed to be found under 'Releases'?
Gabor Szots
CCRL testing group

mhouppin
Posts: 52
Joined: Wed Feb 12, 2020 4:00 pm
Full name: Morgan Houppin

Re: New engine: Stash

Post by mhouppin » Thu Mar 12, 2020 2:07 pm

Fixed, I did not write release notes because I haven't any official binaries yet, I will try to update it this evening.

User avatar
Scally
Posts: 125
Joined: Thu Sep 28, 2017 7:34 pm
Location: Bermondsey, London
Full name: Alan Cooper
Contact:

Re: New engine: Stash

Post by Scally » Mon Mar 16, 2020 12:45 pm

Hi,

I can’t find any official grading for this Engine.

Originally I believe it was estimated around 1450 ELO, but on PicoChess we are seeing v11 holding it’s own against Engines in the 2000-2200 ELO range.


Thanks,

Al.
Alan Cooper
My Chess Computers

Gabor Szots
Posts: 491
Joined: Sat Jul 21, 2018 5:43 am
Location: Szentendre, Hungary
Full name: Gabor Szots

Re: New engine: Stash

Post by Gabor Szots » Mon Mar 16, 2020 1:05 pm

Scally wrote:
Mon Mar 16, 2020 12:45 pm
Hi,

I can’t find any official grading for this Engine.

Originally I believe it was estimated around 1450 ELO, but on PicoChess we are seeing v11 holding it’s own against Engines in the 2000-2200 ELO range.


Thanks,

Al.
You can find a rating for Stash 9.0.1 64-bit in the CCRL blitz list. Version 11 has not been tested yet.
Gabor Szots
CCRL testing group

Post Reply