RomiChess P3n released

[carldaman]

Sorry Mike,
I still do not see how to use these commands correctly. See here the answer of the program inside command prompt:

White<1>learning_on

Error (unknown command): learning_on

r n b q k b n r
p p p p p p p p
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
R N B Q K B N R
hashSig = cb57540b31517d13
hashKey = bd5162a4
pawnKey = 9e8764f7
White<1>uselearn_on

Error (unknown command): uselearn_on

Do you remember the correct commands ?

And how can I uses these commands together inside Arena:

as commandline or as startparameters, and how to separate with ;?

PS: I am using a pgn file for the openings (about 8-10 moves).

best wishes

Hi Werner, Sorry for the confusion. The commands currently in RomiChess are:

learn_on
learn_off
book_on
book_off

They cannot be entered in Arena. So to learn and use the learning enter:

learn_on
book_on
quit

Or just delete the old learn file and Romi will create a new learn file with learning enabled. The values that turn learning on and off are stored in the learn file. Arena and Winboard work fine with Romi's learning but I am not sure about any other.

I would like to thank you for your interest in Romi's learning whether you decide to test with it on or not.

Hi Michael,

Great work with Romi, its style is awesome, even without learning.

Of course, I have learning activated in Romi_P3N, and also merged a few thousand games into its learning file, but a new strange artifact cropped up after I did that: Romi began resigning for no reason a mere 4-6 moves into a game, in even positions. This happened three times, all with white, in a 10 game test match.

I can come up with no explanation for this.

Regards,
CL

[Michael Sherwin]

Sorry Mike,
I still do not see how to use these commands correctly. See here the answer of the program inside command prompt:

White<1>learning_on

Error (unknown command): learning_on

r n b q k b n r
p p p p p p p p
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
R N B Q K B N R
hashSig = cb57540b31517d13
hashKey = bd5162a4
pawnKey = 9e8764f7
White<1>uselearn_on

Error (unknown command): uselearn_on

Do you remember the correct commands ?

And how can I uses these commands together inside Arena:

as commandline or as startparameters, and how to separate with ;?

PS: I am using a pgn file for the openings (about 8-10 moves).

best wishes

Hi Werner, Sorry for the confusion. The commands currently in RomiChess are:

learn_on
learn_off
book_on
book_off

They cannot be entered in Arena. So to learn and use the learning enter:

learn_on
book_on
quit

Or just delete the old learn file and Romi will create a new learn file with learning enabled. The values that turn learning on and off are stored in the learn file. Arena and Winboard work fine with Romi's learning but I am not sure about any other.

I would like to thank you for your interest in Romi's learning whether you decide to test with it on or not.

Hi Michael,

Great work with Romi, its style is awesome, even without learning.

Of course, I have learning activated in Romi_P3N, and also merged a few thousand games into its learning file, but a new strange artifact cropped up after I did that: Romi began resigning for no reason a mere 4-6 moves into a game, in even positions. This happened three times, all with white, in a 10 game test match.

I can come up with no explanation for this.

Regards,
CL

Hi Carl, And thanks again. The problem with Romi's resigning in even positions has something to do with Windows 10 vs 7 & Vista. If I compile it on 10 it won't run at all on Vista/7. I'm using the exact same compiler on both. If I compile on Vista it seems to run perfectly but then that resigning issue showed up on Vista. I can't compile Romi with another compiler unless it supports nameless structs. When compiled on 10 and run on 10 it works flawlessly. When compiled on Vista and run on 10 it works flawlessly.

The kicker is in years past Romi was compiled on Vista with the same compiler, MSVC 2005, and never had that resigning problem. So I am clueless on how to fix that issue.

[carldaman]

Sorry Mike,
I still do not see how to use these commands correctly. See here the answer of the program inside command prompt:

White<1>learning_on

Error (unknown command): learning_on

r n b q k b n r
p p p p p p p p
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
R N B Q K B N R
hashSig = cb57540b31517d13
hashKey = bd5162a4
pawnKey = 9e8764f7
White<1>uselearn_on

Error (unknown command): uselearn_on

Do you remember the correct commands ?

And how can I uses these commands together inside Arena:

as commandline or as startparameters, and how to separate with ;?

PS: I am using a pgn file for the openings (about 8-10 moves).

best wishes

Hi Werner, Sorry for the confusion. The commands currently in RomiChess are:

learn_on
learn_off
book_on
book_off

They cannot be entered in Arena. So to learn and use the learning enter:

learn_on
book_on
quit

Or just delete the old learn file and Romi will create a new learn file with learning enabled. The values that turn learning on and off are stored in the learn file. Arena and Winboard work fine with Romi's learning but I am not sure about any other.

I would like to thank you for your interest in Romi's learning whether you decide to test with it on or not.

Hi Michael,

Great work with Romi, its style is awesome, even without learning.

Of course, I have learning activated in Romi_P3N, and also merged a few thousand games into its learning file, but a new strange artifact cropped up after I did that: Romi began resigning for no reason a mere 4-6 moves into a game, in even positions. This happened three times, all with white, in a 10 game test match.

I can come up with no explanation for this.

Regards,
CL

Hi Carl, And thanks again. The problem with Romi's resigning in even positions has something to do with Windows 10 vs 7 & Vista. If I compile it on 10 it won't run at all on Vista/7. I'm using the exact same compiler on both. If I compile on Vista it seems to run perfectly but then that resigning issue showed up on Vista. I can't compile Romi with another compiler unless it supports nameless structs. When compiled on 10 and run on 10 it works flawlessly. When compiled on Vista and run on 10 it works flawlessly.

The kicker is in years past Romi was compiled on Vista with the same compiler, MSVC 2005, and never had that resigning problem. So I am clueless on how to fix that issue.

Wow, I had no idea it was a compiler/OS issue. The problem occurred on my Win 7 system. I can also try the faster Win 8.1 box to see what happens there. Thanks!

[supersteve3d]

Hi Michael,

I have downloaded Romichess for the first time to see what all the hype is about with the reinforcement learning.

• using RomiChess64P3n.exe
  created as a winboard 2 engine in arena
  learn.dat file autocreated on creation of the engine
  100 game tournament setup with a computer opponent
  tournament is ongoing but learn.dat is still 274KB and not growing

What did I do wrong?

So far Romichess is 0/33

Thanks!

[Damir]

Hi Mike

I posted a link to learning Toga which belongs to the Fruit family. Maybe you can run a sparing match with Romi against it ?Unfortunately the engine keeps crashing. I was wondering if maybe you can take a look at Deep Learning Toga source, and figure out what the cause is behind the crash.

[Guenther]

Hi Michael,

I have downloaded Romichess for the first time to see what all the hype is about with the reinforcement learning.

• using RomiChess64P3n.exe
  created as a winboard 2 engine in arena
  learn.dat file autocreated on creation of the engine
  100 game tournament setup with a computer opponent
  tournament is ongoing but learn.dat is still 274KB and not growing

What did I do wrong?

So far Romichess is 0/33

Thanks!

I guess you tried with an 'ordinary' book with lots of start positions?
In this case try again with a very few selected start positions.
IIRC you need at least some dozens of games per start position for a visible positive effect.

Guenther

[supersteve3d]

Hi Guenther ,

While waiting for reply I changed the tournament to only play the scotch gambit.

I was actually not doubting if I would see a result sooner rather than later.. but rather stating the the learn file seemed to be unchanged. My question was if this was normal. I was expecting the file to grow by a few bytes at least after 30 games or so?

PS. I am also encountering an issue where Romichess resigns once in a while after getting out of book?

Regards,
Steve.

[Guenther]

Hi Guenther ,

While waiting for reply I changed the tournament to only play the scotch gambit.

I was actually not doubting if I would see a result sooner rather than later.. but rather stating the the learn file seemed to be unchanged. My question was if this was normal. I was expecting the file to grow by a few bytes at least after 30 games or so?

PS. I am also encountering an issue where Romichess resigns once in a while after getting out of book?

Regards,
Steve.

IIRC you need a lot of entries for the file to grow.
Did the time stamp change? (changes perhaps only after closing RomiChess)

If you are curious, you could of course compare a fresh learn file with the hopefully updated one,
by looking at the content with a good hex/binary editor, which also reads binary files.
(e.g. HxD for Win: https://mh-nexus.de/de/hxd/)

The resign issue seems known to Michael and is already mentioned
in this thread.

Guenther

[Michael Sherwin]

Hi Michael,

I have downloaded Romichess for the first time to see what all the hype is about with the reinforcement learning.

• using RomiChess64P3n.exe
  created as a winboard 2 engine in arena
  learn.dat file autocreated on creation of the engine
  100 game tournament setup with a computer opponent
  tournament is ongoing but learn.dat is still 274KB and not growing

What did I do wrong?

So far Romichess is 0/33

Thanks!

To see if Romi is learning you can start Romi from the command line and type pr 1 enter. Romi will print out for example, "Record 1 sib 487 chd 2 fs 12 ts 28 some other stuff score 17 w 283 l 264, d 191" which is 1.e2e4. Then type pr 487 in this example to get the record for the first sibling move. In this example it is d2d4, has a score of 19 w 23 l 22 d 13. anyway it is possible to examine the learn file starting with "pr 1".

The resigning problem is new as far as I know. It is an issue between the compiler I use and Windows 7/vista and windows 8/10. The resigning problem does not happen on w10 or as far as I know w8.

[Michael Sherwin]

Hi Mike

I posted a link to learning Toga which belongs to the Fruit family. Maybe you can run a sparing match with Romi against it ?Unfortunately the engine keeps crashing. I was wondering if maybe you can take a look at Deep Learning Toga source, and figure out what the cause is behind the crash.

I would not know where to start. It could be anything anywhere in the code. The master at that kind of bug fixing was Jim Ablett. A day or two ago I looked at Deep Toga's learning and I saw how it worked but today I can't remember what I saw. It was not like the learning in Romi though. Toga with learning would be a moving target for Romi and since they are not close in strength it would take thousands of games for Romi's learning to show.

People need to understand that A0 trained 44 million games. There is a reason for that. A few hundred games can help Romi against equal or close opponents but against multi-threaded SF that is 1000 elo higher it takes a lot more games.