For Spacious Mind

This forum is for general discussions and questions, including Collectors Corner and anything to do with Computer chess.

Moderators: Harvey Williamson, Steve B, Watchman

Forum rules
This textbox is used to restore diagrams posted with the fen tag before the upgrade.
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

For Spacious Mind

Post by Fernando »

May I ask what did you find in that curious program I sent to you?
I mean, about the program itself, the author., etc
Perhaps you found something...

Fern
Festina Lente
User avatar
spacious_mind
Senior Member
Posts: 3999
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Hi Fern

Honestly, I have no idea what to make of it yet.

Alexs (Automatic Learning EXpert System) - 1998

Image

Works on Windows 95, 98 & NT4.0

[Event "Computer Test Match"]
[Site "?"]
[Date "2018.02.17"]
[Round "1"]
[White "AM68060 Sargon 3, 30S."]
[Black "Alexs 2.01, 60/30."]
[Result "1/2-1/2"]
[WhiteElo "2467"]
[BlackElo "2300"]
[PlyCount "82"]
[EventDate "2018.02.17"]
[EventType "match (rapid)"]

1. b3 e5 2. Bb2 Nc6 3. e3 Nf6 4. Bb5 d6 5. Ne2 Bd7 6. O-O {Alexs 2.01 out of book} Ne7 {AM68060 Sargon 3 out of book} 7. Bxd7+ Qxd7 8. f4 Nc6 9. c4 Be7 10. Nbc3 O-O 11. d4 exd4 12. Nxd4 Nxd4 13. Qxd4 c6 14. Rad1 b6 15. Qd3 Rfd8 16. e4 Qe6 17. f5 Qc8 18. Ne2 Nd7 19. Bc3 Nf6 20. Qg3 Qa6 21. Bxf6 Bxf6 22. Rxd6 Qxa2 23. Nc1 Qa1 24. Rfd1 Be7 25. R6d3 Rxd3 26. Qxd3 Qe5 27. Kh1 Bf6 28. Qe3 Re8 29. Re1 Qb2 30. Ne2 Rd8 31. Ng1 Bc3 32. Rc1 Bb4 33. Rf1 Bc3 34. Rc1 Bb4 35. Rf1 Be7 36. e5 Bc5 37. Qg3 Rf8 38. f6 g6 39. Re1 Re8 40. Re2 Qa3 41. Re1 Qb2 {Draw by 3x repetition} 1/2-1/2

Lichess Evaluation

https://lichess.org/ztmAudCv

AM68060 Sargon 3, 30S. (2467)
1 Inaccuracies
2 Mistakes
0 Blunders
18 Average centipawn loss

Alexs 2.01, 60/30. (2300)
4 Inaccuracies
1 Mistakes
0 Blunders
18 Average centipawn loss

I had played a game against AM68080 Sargon 3 and it was drawn by 3x repetition.

I had loaded it onto an MD 1.9 GHz 4 Core computer so it had pretty decent hardware speed. Anyway Sargon 3 was able to draw against it. It seems that Alexs has a pretty decent opening book and understands the middle game. Don't know though yet if it has anything to give in endgames.

I might just include it my future Division 1 Tournament. That will show if it is capable of anything or not.

No idea what they mean with "First Genetic Learning Chess Program".

It says:

"A neural network is a kind of modeling technique that simulates the possibilities of the human brain. A neural network is a kind of new moeling technique that simulates the possibilities of the human brain. Containing billions of neurons the human brain fors the most intelligent system ever known.

A neural Network combines computer power procesing huge amount of data with the ability of the human brain to recognize hidden relations and patterns. Provided with information a human network can train itself!

The program ALEXS can learn to adopt its playing style using neural netowrk guided optimization. It learns by more accurately finding weight factors in the chess program's evaluation function."


Anyway it reads like a load of Gobbelygook to me LMAO. The program itself is fixed 460 KB in size and not growing in size after you played. Neither do you have a learning file.

Therefore no idea what they are talking about other than wanting to sound smart! :P

Best regards
Last edited by spacious_mind on Sat Feb 17, 2018 10:30 pm, edited 1 time in total.
Nick
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

Post by Fernando »

Sounds like almost a hoax but anyway the program play decently well. I wonder if you ever heard of this gentleman and his program. And what then, after...

Kind of a discrete mystery, a lateral way without exit as you find in so many things...

Fern
Festina Lente
User avatar
spacious_mind
Senior Member
Posts: 3999
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Fernando wrote:Sounds like almost a hoax but anyway the program play decently well. I wonder if you ever heard of this gentleman and his program. And what then, after...

Kind of a discrete mystery, a lateral way without exit as you find in so many things...

Fern
You do have a reference here:

https://chessprogramming.wikispaces.com/Alexs

Alex van Tiggelen apparently the author. Funny how the authors name just happens to match exactly ALEXS (Automatic Learning EXpert System)

Nothing ever mentioned since.

Yep I think it is a hoax as well.

Regards
Nick
alynch4047
Member
Posts: 40
Joined: Sat Dec 26, 2015 10:16 am
Location: Bristol

Post by alynch4047 »

I have used neural networks and genetic algorithms for wing optimisation, so they're real things. It sounds very plausible to me,

Abdulhaq
User avatar
spacious_mind
Senior Member
Posts: 3999
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

alynch4047 wrote:I have used neural networks and genetic algorithms for wing optimisation, so they're real things. It sounds very plausible to me,

Abdulhaq
There were lots of discussions going on around the mid/late 90's regarding piece weights/mobility moving away from traditional 1/3/3/5/9 and terms like genetic algorithms were discussed for this. Perhaps it does recalculate its piece values move by move based on pawn structures, safety, space, weak and strong squares, initiative etc... but I doubt that this is any different to what a lot of programmers were developing and including at that time in their chess programs.

Therefore I imagine that there are quite a few programs that evolved with a concept like this.

Alexs was sold in 1998.

What this program claims is that it learns. Therefore if you consider recalculating your piece worth as learning then I guess you can say it does learn as it plays.

Since Alexs was commerically sold, what is misleading is that it does not learn from mistakes and losses it may have made during games in order to not make these same errors again. But it claims it learns "Alexs can learn to adopt its playing style using neural guided network optimization" = piece value adjustment during ONE game. The buyer however reads and expects the program to get better and better through more and more games and it cannot do this!

It can't do this, it has no file that it saves. It's execution file has a fixed non writeable size. No learn files.

Therefore for the person who may have bought this in 1998 this supposedly magical program playing on his 1998 computer would have played weakly and disappointingly as soon as it was tested against the Rebels, Hiarcs, Shredders and Fritz of that day. Plenty of amateur winboard engines would beat it.

So the poor buyer who expected magic would have been disappointed = hoax!

Regards
Nick
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

Post by Fernando »

spacious_mind wrote:
alynch4047 wrote:I have used neural networks and genetic algorithms for wing optimisation, so they're real things. It sounds very plausible to me,

Abdulhaq
There were lots of discussions going on around the mid/late 90's regarding piece weights/mobility moving away from traditional 1/3/3/5/9 and terms like genetic algorithms were discussed for this. Perhaps it does recalculate its piece values move by move based on pawn structures, safety, space, weak and strong squares, initiative etc... but I doubt that this is any different to what a lot of programmers were developing and including at that time in their chess programs.

Therefore I imagine that there are quite a few programs that evolved with a concept like this.

Alexs was sold in 1998.

What this program claims is that it learns. Therefore if you consider recalculating your piece worth as learning then I guess you can say it does learn as it plays.

Since Alexs was commerically sold, what is misleading is that it does not learn from mistakes and losses it may have made during games in order to not make these same errors again. But it claims it learns "Alexs can learn to adopt its playing style using neural guided network optimization" = piece value adjustment during ONE game. The buyer however reads and expects the program to get better and better through more and more games and it cannot do this!

It can't do this, it has no file that it saves. It's execution file has a fixed non writeable size. No learn files.

Therefore for the person who may have bought this in 1998 this supposedly magical program playing on his 1998 computer would have played weakly and disappointingly as soon as it was tested against the Rebels, Hiarcs, Shredders and Fritz of that day. Plenty of amateur winboard engines would beat it.

So the poor buyer who expected magic would have been disappointed = hoax!


BTW, I was not disappointed as much it was a gift from a friend from Serbia in those awful days of the Balkan wars. I do not remember that I ever played it beyond the first moves. Not that I was bother by its strength, probably it is stronger than me, but simply i found boring the GUI and in those days, as in these days, i expect from a program a full immersion with all the frills and whistles you can imagine.
Regards
Festina Lente
User avatar
spacious_mind
Senior Member
Posts: 3999
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Fernando wrote: BTW, I was not disappointed as much it was a gift from a friend from Serbia in those awful days of the Balkan wars. I do not remember that I ever played it beyond the first moves. Not that I was bother by its strength, probably it is stronger than me, but simply i found boring the GUI and in those days, as in these days, i expect from a program a full immersion with all the frills and whistles you can imagine.
Regards
What happened to your friend from Serbia, he used to post here a few years ago and stopped.

Did you know my heritage is Serbian, both my parents are. Difference is that I was born in Germany and grew up in England. As a kid Serbo-Croat was the language that was spoken in our house as that was the only language my parents wanted to talk in the house.

German and later English was learned on the streets and at school :)

ps.... In 1998 the average new computer was a Pentium P2 with 266MHz - 400 MHz speed and 32MB/64MB RAM.

The test game I showed was it playing on an AMD 4 Core at 1.9 GHz with 6 GB RAM. playing against Sargon 3 WinEMU emulator, emulating 68060 Motorola running probably at around 100 MHz Motorola speed.

So yes I would imagine that ALEXS was pretty weak playing on 1998 hardware.

regards
Nick
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

Post by Fernando »

spacious_mind wrote:
Fernando wrote: BTW, I was not disappointed as much it was a gift from a friend from Serbia in those awful days of the Balkan wars. I do not remember that I ever played it beyond the first moves. Not that I was bother by its strength, probably it is stronger than me, but simply i found boring the GUI and in those days, as in these days, i expect from a program a full immersion with all the frills and whistles you can imagine.
Regards
What happened to your friend from Serbia, he used to post here a few years ago and stopped.

Did you know my heritage is Serbian, both my parents are. Difference is that I was born in Germany and grew up in England. As a kid Serbo-Croat was the language that was spoken in our house as that was the only language my parents wanted to talk in the house.

German and later English was learned on the streets and at school :)

regards

I have had not news from him since lot of time. He sent his daughter to my home in order to out her in safety, she was with us several months and then went back home. So this gentleman sent to me lot of things, one of those Alex. I do not have news from her either. I suppose they are OK. I hope they are.
BTW, my bodyguards are Serbian. Tough guys... Imaginary bodyguards of course. My country is 99% quiet and safe, thanks God.
Festina Lente
alynch4047
Member
Posts: 40
Joined: Sat Dec 26, 2015 10:16 am
Location: Bristol

Post by alynch4047 »

spacious_mind wrote:
alynch4047 wrote:I have used neural networks and genetic algorithms for wing optimisation, so they're real things. It sounds very plausible to me,

Abdulhaq
There were lots of discussions going on around the mid/late 90's regarding piece weights/mobility moving away from traditional 1/3/3/5/9 and terms like genetic algorithms were discussed for this. Perhaps it does recalculate its piece values move by move based on pawn structures, safety, space, weak and strong squares, initiative etc... but I doubt that this is any different to what a lot of programmers were developing and including at that time in their chess programs.

Therefore I imagine that there are quite a few programs that evolved with a concept like this.

Alexs was sold in 1998.

What this program claims is that it learns. Therefore if you consider recalculating your piece worth as learning then I guess you can say it does learn as it plays.

Since Alexs was commerically sold, what is misleading is that it does not learn from mistakes and losses it may have made during games in order to not make these same errors again. But it claims it learns "Alexs can learn to adopt its playing style using neural guided network optimization" = piece value adjustment during ONE game. The buyer however reads and expects the program to get better and better through more and more games and it cannot do this!

It can't do this, it has no file that it saves. It's execution file has a fixed non writeable size. No learn files.

Therefore for the person who may have bought this in 1998 this supposedly magical program playing on his 1998 computer would have played weakly and disappointingly as soon as it was tested against the Rebels, Hiarcs, Shredders and Fritz of that day. Plenty of amateur winboard engines would beat it.

So the poor buyer who expected magic would have been disappointed = hoax!

Regards
It does sound very dodgy, but in terms of the techniques discussed genetic algorithms and particularly neural networks work in a completely different way to the standard ways of evaluating best moves. At that time they had no hope against brute force techniques.

Now however neural networks seem to be extremely powerful, ref AlphaGo and AlphaZero.
User avatar
spacious_mind
Senior Member
Posts: 3999
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

alynch4047 wrote: Now however neural networks seem to be extremely powerful, ref AlphaGo and AlphaZero.
Yes, the open question now is will we ever get to see an AlphaZero on our home computers?

OK just to satisfy my curiosity, I will set up a game between Rebel 10 (1998) playing in DOSBox with 200,000 CPU cycles = Pentium P2 300 MHz (approx.) against ALEXS on my I7-7700HQ 2.8 GHz 12GB RAM laptop and see what happens.

Regards
Nick
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

Post by Fernando »

spacious_mind wrote:
alynch4047 wrote: Now however neural networks seem to be extremely powerful, ref AlphaGo and AlphaZero.
Yes, the open question now is will we ever get to see an AlphaZero on our home computers?

OK just to satisfy my curiosity, I will set up a game between Rebel 10 (1998) playing in DOSBox with 200,000 CPU cycles = Pentium P2 300 MHz (approx.) against ALEXS on my I7-7700HQ 2.8 GHz 12GB RAM laptop and see what happens.

Regards
My guess: Rebel will maul the poor creature.
Festina Lente
User avatar
spacious_mind
Senior Member
Posts: 3999
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Fernando wrote: My guess: Rebel will maul the poor creature.
Well you got that right :)

Lichess Evaluation

https://lichess.org/BUfTVjlI

DB P2-300 MHz - Rebel 10, 30S.
1 Inaccuracies
0 Mistakes
0 Blunders
13 Average centipawn loss

I7-2.8 GHz Alexs 2.01, 60/30.
6 Inaccuracies
3 Mistakes
1 Blunders
33 Average centipawn loss

[Event "Computer Test Match"]
[Site "Pelham, AL"]
[Date "2018.02.18"]
[Round "?"]
[White "DB P2-300 MHz - Rebel 10, 30S."]
[Black "I7-2.8 GHz Alexs 2.01, 60/30."]
[Result "1-0"]
[PlyCount "97"]
[EventDate "2018.02.18"]
[EventCountry "USA"]

1. d4 d5 2. c4 dxc4 3. e3 Nf6 4. Nf3 e6 5. Bxc4 c5 6. O-O Nc6 7. Qe2 a6 8. dxc5 Bxc5 9. e4 b5 10. Bb3 Bb7 11. Nc3 {I7-2.8 GHz Alexs 2.01 out of book} b4 {DB P2-300 MHz - Rebel 10 out of book} 12. Rd1 Qe7 13. Na4 Ba7 14. Bg5 h6 15. Bxf6 Qxf6 16. Rac1 Qe7 17. Qc2 O-O 18. Nc5 Qxc5 19. Qxc5 Bxc5 20. Rxc5 Rad8 21. Rdc1 Nd4 22. Nxd4 Rxd4 23. f3 Rd2 24. Rc7 Rb8 25. Re7 Rxb2 26. Rcc7 Bxe4 27. fxe4 Rb1+ 28. Kf2 Rb2+ 29. Kg3 Rd8 30. Rxf7 Rd3+ 31. Rf3 Rdd2 32. Bxe6+ Kh7 33. Bh3 Kg8 34. e5 Rd8 35. Rff7 Kh8 36. e6 Rg8 37. e7 Re2 38. Bf5 b3 39. Rc8 g6 40. Rff8 Kh7 41. Rxg8 gxf5 42. e8=Q Rxe8 43. Rgxe8 Kg7 44. Rc6 Kf7 45. Ra8 Ke7 46. Ra7+ Kd8 47. Rxh6 f4+ 48. Kxf4 bxa2 49. Rh8# 1-0

Final Position

[fen]3k3R/R7/p7/8/5K2/8/p5PP/8 w - - 0 49[/fen]

Well that was as much hardware speed as what I could give it. So I can imagine someone cussing after having just bought it in 1998 :)

Anyway even if the idea is good, my instincts tell me that you have to have a really good program to begin with that you experiment these principles on. In the above game example I would imagine if it were tried with Rebel 10, then perhaps Rebel 10 would lose calculation speed but still it would be more successful at implementing genetic algorithms and neural networks than the base program used by ALEXS because Rebel 10 has much better chess knowledge.

It kind of reinforces my thoughts that even Alpha Zero would be hopeless unless it started with a sold excellent base chess program. Therefore my opinion is Alpha Zero is open source Stockfish around which neural network principles were added.

It's probably also the reason why Stockfish could only at best draw against it, since it was playing an advanced version of itself.

Anyway those are my food for thoughts....

Regards
Nick
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

Post by Fernando »

I was taking a look at the other stuff that sent to me this guy from Serbia, Dvidanovic is his name. There are some opening books by CB of those years and a CD bought in a Moscow market place full of engines of those years. Pirate stuff for certain. Probably today already available anyway in abadonware places. I will take a look at it just in case there is something of interest for you.

Fern
Festina Lente
User avatar
Fernando
Admiral of the Fleet
Posts: 3059
Joined: Tue Jul 31, 2007 4:35 pm
Location: Santiago de Chile

Post by Fernando »

These are the programs embedded in the CD that MAYBE ARE NOT available in Internet places:

CM 5500
VChess


And nothing more. Ther rest is these day commodity stuff everywhere.
Festina Lente
Post Reply