mcostalba wrote:I have read the paper: result is impressive!
Honestly I didn't think it was possible because my understanding was that chess is more "computer friendly" than Go....I was wrong.
It is true, SF is not meant to play at its best without a book and especially 1 fixed minute per move cuts out the whole time management, it would be more natural to play with tournament conditions, but nevertheless I think these are secondary aspects, what has been accomplished is huge.
Marco, A0 did not win a match against SF. A0 with RL won a match against SF. Or said another way, A0 won a match against SF because SF does not have RL. Or thought of a different way, a group of programmers identified a deficiency that exist in a competitive field and took advantage of that deficiency by eliminating that deficiency in their entity. Or one can change that thought around and say RL does not belong in competitive chess because it covers up the underlying strength and correctness of the algorithm. In that case the A0 vs SF match is non sequitur and meaningless. Then there is the thought of the fan that wants RL but are ignored because they are not important and what the fan thinks or wants is not meaningful.
But, what you can't say is, "what has been accomplished is huge" in terms of a chess playing algorithm. You might say that what A0 has demonstrated in go, chess and shogi has accomplished a huge demonstration that a NN with RL may conquer hamanity some day. I won't argue against that. Concerning chess though the AB algorithm is not inferior to NN+MC. It is inferior to NN+MC+RL. AB+RL is far superior to NN+MC+RL.
And I said all that without mentioning RomiChess not even one time!
That alpha-beta search + reinforcement learning is indeed better than mcts + nn+reinforcement learning is still something that has to be proven. Assertions and a bulk of text doesn't help it Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out
mcostalba wrote:I have read the paper: result is impressive!
Honestly I didn't think it was possible because my understanding was that chess is more "computer friendly" than Go....I was wrong.
It is true, SF is not meant to play at its best without a book and especially 1 fixed minute per move cuts out the whole time management, it would be more natural to play with tournament conditions, but nevertheless I think these are secondary aspects, what has been accomplished is huge.
Marco, A0 did not win a match against SF. A0 with RL won a match against SF. Or said another way, A0 won a match against SF because SF does not have RL. Or thought of a different way, a group of programmers identified a deficiency that exist in a competitive field and took advantage of that deficiency by eliminating that deficiency in their entity. Or one can change that thought around and say RL does not belong in competitive chess because it covers up the underlying strength and correctness of the algorithm. In that case the A0 vs SF match is non sequitur and meaningless. Then there is the thought of the fan that wants RL but are ignored because they are not important and what the fan thinks or wants is not meaningful.
But, what you can't say is, "what has been accomplished is huge" in terms of a chess playing algorithm. You might say that what A0 has demonstrated in go, chess and shogi has accomplished a huge demonstration that a NN with RL may conquer hamanity some day. I won't argue against that. Concerning chess though the AB algorithm is not inferior to NN+MC. It is inferior to NN+MC+RL. AB+RL is far superior to NN+MC+RL.
And I said all that without mentioning RomiChess not even one time!
That alpha-beta search + reinforcement learning is indeed better than mcts + nn+reinforcement learning is still something that has to be proven. Assertions and a bulk of text doesn't help it Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out
Technically correct but not practically correct. Demonstrably there is strong evidence supporting what I posted. It was demonstrated by R_m_C_e_s_ that hundreds of elo can be gained just by very few training games in real competition. And over a 1000 elo in very restrictive test with even fewer training games. Against a truly massive opening book and against 6 top engines it was demonstrated that 50 elo per 5,000 games of training is achieved. And the gain was linear during the scope of the test. So unless it is believed that a 2400 elo engine can benefit this way but a 3400 elo engine cannot then it can be assumed that the 3400 elo engine will do quite well. In the case of SF that would mean victory against A0.
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through
mcostalba wrote:I have read the paper: result is impressive!
Honestly I didn't think it was possible because my understanding was that chess is more "computer friendly" than Go....I was wrong.
It is true, SF is not meant to play at its best without a book and especially 1 fixed minute per move cuts out the whole time management, it would be more natural to play with tournament conditions, but nevertheless I think these are secondary aspects, what has been accomplished is huge.
Marco, A0 did not win a match against SF. A0 with RL won a match against SF. Or said another way, A0 won a match against SF because SF does not have RL. Or thought of a different way, a group of programmers identified a deficiency that exist in a competitive field and took advantage of that deficiency by eliminating that deficiency in their entity. Or one can change that thought around and say RL does not belong in competitive chess because it covers up the underlying strength and correctness of the algorithm. In that case the A0 vs SF match is non sequitur and meaningless. Then there is the thought of the fan that wants RL but are ignored because they are not important and what the fan thinks or wants is not meaningful.
But, what you can't say is, "what has been accomplished is huge" in terms of a chess playing algorithm. You might say that what A0 has demonstrated in go, chess and shogi has accomplished a huge demonstration that a NN with RL may conquer hamanity some day. I won't argue against that. Concerning chess though the AB algorithm is not inferior to NN+MC. It is inferior to NN+MC+RL. AB+RL is far superior to NN+MC+RL.
And I said all that without mentioning RomiChess not even one time!
That alpha-beta search + reinforcement learning is indeed better than mcts + nn+reinforcement learning is still something that has to be proven. Assertions and a bulk of text doesn't help it Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out
Technically correct but not practically correct. Demonstrably there is strong evidence supporting what I posted. It was demonstrated by R_m_C_e_s_ that hundreds of elo can be gained just by very few training games in real competition. And over a 1000 elo in very restrictive test with even fewer training games. Against a truly massive opening book and against 6 top engines it was demonstrated that 50 elo per 5,000 games of training is achieved. And the gain was linear during the scope of the test. So unless it is believed that a 2400 elo engine can benefit this way but a 3400 elo engine cannot then it can be assumed that the 3400 elo engine will do quite well. In the case of SF that would mean victory against A0.
Can you point to a link ? Where there a sufficient amount of test games played ? Would like to see the statistics
mcostalba wrote:I have read the paper: result is impressive!
Honestly I didn't think it was possible because my understanding was that chess is more "computer friendly" than Go....I was wrong.
It is true, SF is not meant to play at its best without a book and especially 1 fixed minute per move cuts out the whole time management, it would be more natural to play with tournament conditions, but nevertheless I think these are secondary aspects, what has been accomplished is huge.
Marco, A0 did not win a match against SF. A0 with RL won a match against SF. Or said another way, A0 won a match against SF because SF does not have RL. Or thought of a different way, a group of programmers identified a deficiency that exist in a competitive field and took advantage of that deficiency by eliminating that deficiency in their entity. Or one can change that thought around and say RL does not belong in competitive chess because it covers up the underlying strength and correctness of the algorithm. In that case the A0 vs SF match is non sequitur and meaningless. Then there is the thought of the fan that wants RL but are ignored because they are not important and what the fan thinks or wants is not meaningful.
But, what you can't say is, "what has been accomplished is huge" in terms of a chess playing algorithm. You might say that what A0 has demonstrated in go, chess and shogi has accomplished a huge demonstration that a NN with RL may conquer hamanity some day. I won't argue against that. Concerning chess though the AB algorithm is not inferior to NN+MC. It is inferior to NN+MC+RL. AB+RL is far superior to NN+MC+RL.
And I said all that without mentioning RomiChess not even one time!
That alpha-beta search + reinforcement learning is indeed better than mcts + nn+reinforcement learning is still something that has to be proven. Assertions and a bulk of text doesn't help it Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out
Technically correct but not practically correct. Demonstrably there is strong evidence supporting what I posted. It was demonstrated by R_m_C_e_s_ that hundreds of elo can be gained just by very few training games in real competition. And over a 1000 elo in very restrictive test with even fewer training games. Against a truly massive opening book and against 6 top engines it was demonstrated that 50 elo per 5,000 games of training is achieved. And the gain was linear during the scope of the test. So unless it is believed that a 2400 elo engine can benefit this way but a 3400 elo engine cannot then it can be assumed that the 3400 elo engine will do quite well. In the case of SF that would mean victory against A0.
Can you point to a link ? Where there a sufficient amount of test games played ? Would like to see the statistics
These test were done over ten years ago. The test against a massive opening book and 6 top engines was done by Marc Lacrosse and he has passed away. There are people still around that remember those test. Maybe they can help. The point is that this was all demonstrated a decade ago and very few people took notice. The overall bias was against it. A few fans of Romi's learning have begged authors of the top engines for over a decade to include Romi style learning. They would pay money for it. But for over ten years they have been ignored. A0 may change that. Or it may not.
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through
mcostalba wrote:I have read the paper: result is impressive!
Honestly I didn't think it was possible because my understanding was that chess is more "computer friendly" than Go....I was wrong.
It is true, SF is not meant to play at its best without a book and especially 1 fixed minute per move cuts out the whole time management, it would be more natural to play with tournament conditions, but nevertheless I think these are secondary aspects, what has been accomplished is huge.
Marco, A0 did not win a match against SF. A0 with RL won a match against SF. Or said another way, A0 won a match against SF because SF does not have RL. Or thought of a different way, a group of programmers identified a deficiency that exist in a competitive field and took advantage of that deficiency by eliminating that deficiency in their entity. Or one can change that thought around and say RL does not belong in competitive chess because it covers up the underlying strength and correctness of the algorithm. In that case the A0 vs SF match is non sequitur and meaningless. Then there is the thought of the fan that wants RL but are ignored because they are not important and what the fan thinks or wants is not meaningful.
But, what you can't say is, "what has been accomplished is huge" in terms of a chess playing algorithm. You might say that what A0 has demonstrated in go, chess and shogi has accomplished a huge demonstration that a NN with RL may conquer hamanity some day. I won't argue against that. Concerning chess though the AB algorithm is not inferior to NN+MC. It is inferior to NN+MC+RL. AB+RL is far superior to NN+MC+RL.
And I said all that without mentioning RomiChess not even one time!
That alpha-beta search + reinforcement learning is indeed better than mcts + nn+reinforcement learning is still something that has to be proven. Assertions and a bulk of text doesn't help it Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out
Technically correct but not practically correct. Demonstrably there is strong evidence supporting what I posted. It was demonstrated by R_m_C_e_s_ that hundreds of elo can be gained just by very few training games in real competition. And over a 1000 elo in very restrictive test with even fewer training games. Against a truly massive opening book and against 6 top engines it was demonstrated that 50 elo per 5,000 games of training is achieved. And the gain was linear during the scope of the test. So unless it is believed that a 2400 elo engine can benefit this way but a 3400 elo engine cannot then it can be assumed that the 3400 elo engine will do quite well. In the case of SF that would mean victory against A0.
Can you point to a link ? Where there a sufficient amount of test games played ? Would like to see the statistics
These test were done over ten years ago. The test against a massive opening book and 6 top engines was done by Marc Lacrosse and he has passed away. There are people still around that remember those test. Maybe they can help. The point is that this was all demonstrated a decade ago and very few people took notice. The overall bias was against it. A few fans of Romi's learning have begged authors of the top engines for over a decade to include Romi style learning. They would pay money for it. But for over ten years they have been ignored. A0 may change that. Or it may not.
The problem is that if you can not provide me with the actual data the discussion pretty much ends there I am not saying that there isn't any data but until you have provided it, I can not take your argument seriously. "Some people did some tests 10 years ago" wouldn't hold up in court and definently not in a scientific discussion.
mcostalba wrote:I have read the paper: result is impressive!
Honestly I didn't think it was possible because my understanding was that chess is more "computer friendly" than Go....I was wrong.
It is true, SF is not meant to play at its best without a book and especially 1 fixed minute per move cuts out the whole time management, it would be more natural to play with tournament conditions, but nevertheless I think these are secondary aspects, what has been accomplished is huge.
Marco, A0 did not win a match against SF. A0 with RL won a match against SF. Or said another way, A0 won a match against SF because SF does not have RL. Or thought of a different way, a group of programmers identified a deficiency that exist in a competitive field and took advantage of that deficiency by eliminating that deficiency in their entity. Or one can change that thought around and say RL does not belong in competitive chess because it covers up the underlying strength and correctness of the algorithm. In that case the A0 vs SF match is non sequitur and meaningless. Then there is the thought of the fan that wants RL but are ignored because they are not important and what the fan thinks or wants is not meaningful.
But, what you can't say is, "what has been accomplished is huge" in terms of a chess playing algorithm. You might say that what A0 has demonstrated in go, chess and shogi has accomplished a huge demonstration that a NN with RL may conquer hamanity some day. I won't argue against that. Concerning chess though the AB algorithm is not inferior to NN+MC. It is inferior to NN+MC+RL. AB+RL is far superior to NN+MC+RL.
And I said all that without mentioning RomiChess not even one time!
That alpha-beta search + reinforcement learning is indeed better than mcts + nn+reinforcement learning is still something that has to be proven. Assertions and a bulk of text doesn't help it Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out
Technically correct but not practically correct. Demonstrably there is strong evidence supporting what I posted. It was demonstrated by R_m_C_e_s_ that hundreds of elo can be gained just by very few training games in real competition. And over a 1000 elo in very restrictive test with even fewer training games. Against a truly massive opening book and against 6 top engines it was demonstrated that 50 elo per 5,000 games of training is achieved. And the gain was linear during the scope of the test. So unless it is believed that a 2400 elo engine can benefit this way but a 3400 elo engine cannot then it can be assumed that the 3400 elo engine will do quite well. In the case of SF that would mean victory against A0.
Can you point to a link ? Where there a sufficient amount of test games played ? Would like to see the statistics
These test were done over ten years ago. The test against a massive opening book and 6 top engines was done by Marc Lacrosse and he has passed away. There are people still around that remember those test. Maybe they can help. The point is that this was all demonstrated a decade ago and very few people took notice. The overall bias was against it. A few fans of Romi's learning have begged authors of the top engines for over a decade to include Romi style learning. They would pay money for it. But for over ten years they have been ignored. A0 may change that. Or it may not.
The problem is that if you can not provide me with the actual data the discussion pretty much ends there I am not saying that there isn't any data but until you have provided it, I can not take your argument seriously. "Some people did some tests 10 years ago" wouldn't hold up in court and definently not in a scientific discussion.
I can't care less if you take my argument seriously. There are people still here from 10 years ago that have backed me up and can back me up if they chose to do so but they would have to put up with people like you and that is not easy to do.
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through
That's just completly garbage. Either you provide the data or you dont. Until you or anyone else does there is no point in taking you seriously. "As for dealing with ppl like me". People who like to look at the actual data ? What's wrong with that ?
CheckersGuy wrote:That's just completly garbage. Either you provide the data or you dont. Until you or anyone else does there is no point in taking you seriously. "As for dealing with ppl like me". People who like to look at the actual data ? What's wrong with that ?
It was not your quest it was your attitude. It is obtrusive. Just look at the Chess Programming Wiki for RomiChess and you will see that it is written that RomiChess is famous for its learning. Do you think Gerd Isenberg was just saying that for no good reason? There are links to post about Romi's learning there.
But be honest, you were opposed to my claims from the beginning. I read your post. I don't believe that you are interested in Romi's learning or any 10 year old saved data. But it is so very convenient for you to ask for 10+ year old data and dismiss my claims because I chose not to produce or can't produce. No! Then prove me wrong and do a little research. Or better yet generate your own data. If you care you will do that because it would be superior to third party knowledge. But I reiterate that you do not care and are playing a game of one upmanship only.
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through
CheckersGuy wrote:That's just completly garbage. Either you provide the data or you dont. Until you or anyone else does there is no point in taking you seriously. "As for dealing with ppl like me". People who like to look at the actual data ? What's wrong with that ?
It was not your quest it was your attitude. It is obtrusive. Just look at the Chess Programming Wiki for RomiChess and you will see that it is written that RomiChess is famous for its learning. Do you think Gerd Isenberg was just saying that for no good reason? There are links to post about Romi's learning there.
But be honest, you were opposed to my claims from the beginning. I read your post. I don't believe that you are interested in Romi's learning or any 10 year old saved data. But it is so very convenient for you to ask for 10+ year old data and dismiss my claims because I chose not to produce or can't produce. No! Then prove me wrong and do a little research. Or better yet generate your own data. If you care you will do that because it would be superior to third party knowledge. But I reiterate that you do not care and are playing a game of one upmanship only.
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through
CheckersGuy wrote:That's just completly garbage. Either you provide the data or you dont. Until you or anyone else does there is no point in taking you seriously. "As for dealing with ppl like me". People who like to look at the actual data ? What's wrong with that ?
It was not your quest it was your attitude. It is obtrusive. Just look at the Chess Programming Wiki for RomiChess and you will see that it is written that RomiChess is famous for its learning. Do you think Gerd Isenberg was just saying that for no good reason? There are links to post about Romi's learning there.
But be honest, you were opposed to my claims from the beginning. I read your post. I don't believe that you are interested in Romi's learning or any 10 year old saved data. But it is so very convenient for you to ask for 10+ year old data and dismiss my claims because I chose not to produce or can't produce. No! Then prove me wrong and do a little research. Or better yet generate your own data. If you care you will do that because it would be superior to third party knowledge. But I reiterate that you do not care and are playing a game of one upmanship only.
I have been extremely impressed with romi chess' learning ability. I have inserted michael sherwin's explanation of it below. If you focus your games on a single position, it leads to considerable improvement in playing strength (With learning on, and lots of training sessions, romi is beating engines over 200 points stronger than itself).
It would be great if rybka could implement something like this. It would find the absolute best lines eventually. And it would be even better if you could see the conclusions of the learning in tree form. (e.g., this line gets a -.2 penality, this line gets +.3)
I have inserted a description of the learning at the very bottom
Here is a tournement i've been running from the position 1 e4, c6; 2; d4, d5; 3; c4!? You can see romi improving against these better engines. It is often getting 40 moves deep before it has to think. I will keep adding stronger and stronger engines, to make sure romi finds solutions that apply across the board. It has just started beating list, so next up is ruffian. I wonder how strong an engine i can put against it before it stops improving.
quote from michael sherwin, describing romi's learning,
"here were lots of dicussions about how Romi's learning works. Maybe someone will provide the links. I only have time for a short explaination, right now. I will write something up to include with my next release.
The 'skinny' is, Romi combines two ideas in her learn file, 1)Pavlov's dog experiment and 2)monkey see monkey do.
First, Romi remembers every move of every game for 160 ply (IIRC) and
1) gives a small bonus to every move of the winning side and a small penalty to every move of the loosing side (Pavlov)
2) Romi plays back winning lines played by herself as well as those played by her opponents (Monkey see)
so, Romi will without 'thinking' play back your own lines against you if you are good enough to beat her or her lines untill you beat her at wich time the bonus/penalty will kick in to make Romi change her play untill she finds a way to win."
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through
Only a match between engines using those 2 different algorithms can be thought of as an definitive answer. Everything else is just based on certain assumptions. Since we won't have a commercial version of AlphaZero anytime soon it probably will be quite some time until we find out 