Page 8 of 9

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 11:39 am
by Rodolfo Leoni
Michael Sherwin wrote: .........................................................
Romi's learning is a bit more than position learning. When Romi learns nodes higher in the tree are affected the most and change sooner. However, as more results come in the moves at the root get better defined. So for example Romi will choose between 1.e4 and 1.d4 which ever gives Romi a better result. That is true from any node in the tree. That is a permanent gain. It may not help win matches against god engines but it will help Romi gain several classes in strength against her contemporaries as demonstrated in Leo's class tournaments where Romi gained two classes and was about to gain a third. And that was based on just a 100 to 200 played games!
The more I read about AlphaZ, the more I think they adapted your ideas to build a learning software on a divine hardware...

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 12:15 pm
by Michael Sherwin
Rodolfo Leoni wrote:
Michael Sherwin wrote: .........................................................
Romi's learning is a bit more than position learning. When Romi learns nodes higher in the tree are affected the most and change sooner. However, as more results come in the moves at the root get better defined. So for example Romi will choose between 1.e4 and 1.d4 which ever gives Romi a better result. That is true from any node in the tree. That is a permanent gain. It may not help win matches against god engines but it will help Romi gain several classes in strength against her contemporaries as demonstrated in Leo's class tournaments where Romi gained two classes and was about to gain a third. And that was based on just a 100 to 200 played games!
The more I read about AlphaZ, the more I think they adapted your ideas to build a learning software on a divine hardware...
I think it is a 100% possibility that the AlphaZ team is at least aware of RomiChess's learning algorithm. Especially with a quote like the following from the chess programming wiki.

"RomiChess is famous for its learning approach"

And the fact that at least one individual on the AlphaZ team is a member of Talkchess. This is circumstantial evidence albeit strong circumstantial evidence but not quite proof. However, if any prior endeavor helped to show the way it is a very strong probability RomiChess had a part in that. If so it would be nice if Google-Deep-Mind gave an honorary mention to RomiChess and its author.

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 12:29 pm
by Rodolfo Leoni
Michael Sherwin wrote:
Rodolfo Leoni wrote:
Michael Sherwin wrote: .........................................................
Romi's learning is a bit more than position learning. When Romi learns nodes higher in the tree are affected the most and change sooner. However, as more results come in the moves at the root get better defined. So for example Romi will choose between 1.e4 and 1.d4 which ever gives Romi a better result. That is true from any node in the tree. That is a permanent gain. It may not help win matches against god engines but it will help Romi gain several classes in strength against her contemporaries as demonstrated in Leo's class tournaments where Romi gained two classes and was about to gain a third. And that was based on just a 100 to 200 played games!
The more I read about AlphaZ, the more I think they adapted your ideas to build a learning software on a divine hardware...
I think it is a 100% possibility that the AlphaZ team is at least aware of RomiChess's learning algorithm. Especially with a quote like the following from the chess programming wiki.

"RomiChess is famous for its learning approach"

And the fact that at least one individual on the AlphaZ team is a member of Talkchess. This is circumstantial evidence albeit strong circumstantial evidence but not quite proof. However, if any prior endeavor helped to show the way it is a very strong probability RomiChess had a part in that. If so it would be nice if Google-Deep-Mind gave an honorary mention to RomiChess and its author.
Agreed 100%.

Every time I read about space explorations I think at Jules Verne's books... :D

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 3:14 pm
by Uri Blass
syzygy wrote:
Rodolfo Leoni wrote:One thing is 100% sure: when Google hardware will be available then correspondence chess will die. What's the point in playing games if the learning feature is advanced at a point where it only plays perfect games?
There is no reason to think that AlphaZero plays only perfect games. It might not even know about the wrong corner in KBNK. (Does that come up often enough in 44 million games for AlphaZero to figure it out? Maybe not, maybe it does, but even if it does there will be other patterns it won't have seen often enough.)

Show it one of those positions that are easy for humans, that fool all top engines and that never come up in real games, and I guarantee you AlphaZero will not have a clue either.
I believe that after enough time alphazero will learn.
In the same way that it learned about the opening position it can learn every new position.

If a program need some minutes to find the solution in positions that are easy for humans by learning by playing against itself then it is still better than the programs of today that practically cannnot do it it by brute force even after some hours.

Note that KBN vs K is a bad example because programs without bugs should find the mate by search even without a special knowledge about the right corner and without tablebases but it is possible to find some position when search does not help.


Stockfish without tablabases and without special knowledge can easily see a winning score


[D]N7/8/8/8/3k4/8/8/K6B w - - 0 1

Stockfish_17101111_x64_modern:
1/1 00:00 44 44k +42.54 1.Kb2
2/2 00:00 148 148k +42.45 1.Kb2 Ke5
3/3 00:00 266 266k +42.54 1.Kb2 Ke5 2.Kc3
4/4 00:00 642 321k +42.54 1.Kb2 Kd3 2.Nb6 Kd4
5/5 00:00 1k 502k +42.54 1.Kb2 Kd3 2.Nb6 Kd4 3.Kc2
6/6 00:00 2k 874k +42.54 1.Kb2 Kd3 2.Nb6 Kd4 3.Na4 Ke5
7/8 00:00 4k 1,913k +42.60 1.Kb2 Kc4 2.Kc2 Kd4 3.Kd2 Ke5 4.Kc3
8/8 00:00 6k 2,044k +42.58 1.Kb2 Kc4 2.Kc2 Kd4 3.Kd2 Ke5 4.Kc3 Kf6
9/10 00:00 12k 3,098k +42.54 1.Kb2 Kc4 2.Kc2 Kd4 3.Kd2 Kc4 4.Nc7 Kd4 5.Nd5
10/13 00:00 16k 3,234k +42.62 1.Kb2 Ke5 2.Kc3 Kf6 3.Kd4 Kf5 4.Bd5 Kf6 5.Ke4 Kg7 6.Kf5
11/13 00:00 20k 3,382k +42.62 1.Kb2 Ke5 2.Kc3 Kf6 3.Kd4 Kf5 4.Bd5 Kf6 5.Ke4 Kg7 6.Kf5 Kh8 7.Kf6
12/14 00:00 28k 3,984k +42.62 1.Kb2 Ke5 2.Kc3 Kf6 3.Kd4 Kf5 4.Bd5 Kg6 5.Ke5 Kg7 6.Kf5 Kh7 7.Kf6 Kh8
13/16 00:00 37k 4,072k +42.62 1.Kb2 Ke5 2.Kc3 Kf6 3.Kd4 Kf5 4.Bd5 Kf6 5.Ke4 Kg5 6.Ke5 Kg6 7.Bh1 Kg7 8.Kf5
14/16 00:00 68k 4,012k +42.62 1.Kb2 Kc4 2.Kc2 Kb4 3.Kd2 Kc4 4.Nc7 Kd4 5.Nd5 Ke5 6.Kd3 Kf5 7.Kd4 Kg6 8.Ke5
15/20 00:00 93k 4,227k +42.62 1.Kb2 Kc4 2.Kc2 Kd4 3.Kd2 Kc4 4.Nc7 Kd4 5.Nd5 Ke5 6.Kd3 Kf5 7.Kd4 Kg6 8.Ke5 Kf7 9.Nb4 Kg7
16/18 00:00 105k 4,366k +42.62 1.Kb2 Kc4 2.Kc2 Kd4 3.Kd2 Kc4 4.Nc7 Kd4 5.Nd5 Ke5 6.Kd3 Kf5 7.Kd4 Kg6 8.Ke5 Kf7 9.Nb4 Kg7
17/19 00:00 137k 4,410k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Be4 Kf7 9.Bb1
18/18 00:00 147k 4,332k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Be4 Kf7 9.Bc6 Kg8
19/19 00:00 193k 4,014k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Be4 Kf7 9.Bc6 Kg8 10.Kf6
20/21 00:00 212k 4,073k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Be4 Kf7 9.Bb1 Kg7 10.Ba2 Kh8
21/22 00:00 240k 4,204k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Be4 Kf7 9.Bc6 Kg8 10.Bd5+ Kh8 11.Kf6 Kh7
22/23 00:00 280k 4,176k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Be4 Kf7 9.Bc6 Kg8 10.Kf6 Kh8 11.Bd5 Kh7
23/24 00:00 330k 4,292k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Ke5 Kh8 7.Kf6 Kg8 8.Ne6 Kh8 9.Kf7 Kh7 10.Be4+ Kh8 11.Bc6 Kh7 12.Nf4 Kh8
24/25 00:00 382k 4,336k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Nc7 Kh8 9.Kf6 Kg8 10.Ne6 Kh8 11.Nd8 Kg8 12.Nf7 Kh7
25/27 00:00 478k 4,422k +42.66 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kg7 6.Na8 Kf8 7.Ke5 Kg7 8.Nc7 Kh8 9.Kf6 Kg8 10.Ne6 Kh8 11.Nd8 Kg8 12.Nf7 Kh7 13.Bd5
26/29 00:00 733k 4,413k +42.70 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Na8 Kf6 7.Bd5 Kg7 8.Ke5 Kg6 9.Nc7 Kh5 10.Be4 Kg4 11.Ne6 Kh4 12.Kf6 Kg4 13.Kg7 Kg3
27/28 00:00 840k 4,467k +42.70 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Na8 Kf6 7.Bd5 Kf5 8.Nc7 Kg5 9.Ke5 Kg4 10.Be4 Kg3 11.Bc2 Kg4 12.Bf5+ Kf3 13.Nd5 Ke2 14.Kf4 Kd2
28/28 00:00 1,018k 4,525k +42.70 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Na8 Kf6 7.Bd5 Kf5 8.Nc7 Kf4 9.Be4 Kg4 10.Ke5 Kg3 11.Nd5 Kf2 12.Kd4 Kg3 13.Bf5 Kf2 14.Bg6 Ke1
29/37 00:00 2,271k 5,369k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Bh1 Kg7 16.Be4
30/33 00:00 2,395k 5,357k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Bh1 Kg7 16.Kd4 Kg8 17.Bd5+ Kh8
31/36 00:00 2,643k 5,427k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Bh1 Kg7 16.Kd4 Kg8 17.Bd5+ Kh8 18.Ke5 Kg7
32/36 00:00 2,875k 5,486k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Kf6 Kg8 16.Bh1 Kh7 17.Be4+ Kh8 18.Nb5 Kg8
33/40 00:00 3,019k 5,479k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Kf6 Kg8 16.Bh1 Kh7 17.Be4+ Kh8 18.Ke5 Kg8 19.Nb5 Kf7 20.Bd5+ Kg6
34/39 00:00 3,378k 5,484k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Kf6 Kg8 16.Bh1 Kh7 17.Ke5 Kg8 18.Kd4 Kh8 19.Nb5 Kg7
35/40 00:00 3,680k 5,534k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Kf6 Kg8 16.Bh1 Kh7 17.Ke5 Kg8 18.Na8 Kg7 19.Kd4 Kh6 20.Ke5
36/41 00:00 4,220k 5,635k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Na8 Kg7 16.Bd5 Kf8 17.Kf6 Ke8 18.Nc7+ Kf8 19.Nb5 Ke8 20.Bf3 Kf8
37/42 00:00 4,609k 5,690k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kf7 14.Ke5 Kf8 15.Na8 Kg8 16.Bd5+ Kh8 17.Kf6 Kh7 18.Nc7 Kh8 19.Bh1 Kg8 20.Ke5 Kf7 21.Na8 Kg7
38/45 00:00 5,356k 5,803k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kf8 15.Kf6 Kg8 16.Bd5+ Kh8 17.Ke5 Kh7 18.Bf3 Kh6 19.Bh1 Kh5 20.Na8 Kh4 21.Kf4 Kh5
39/42 00:01 6,039k 5,785k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kf4 8.Ne8 Kg4 9.Ke3 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kf8 15.Na8 Kg7 16.Kd4 Kg6 17.Be4+ Kg7 18.Ke5 Kh8 19.Bh1 Kg7 20.Bf3 Kh7 21.Kf6 Kh8
40/42 00:01 6,946k 5,750k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh7 15.Na8 Kg7 16.Bh1 Kf7 17.Bd5+ Kg6 18.Be4+ Kg7 19.Nc7 Kh6 20.Bh1 Kh5 21.Kf4 Kg6
41/43 00:01 8,051k 5,730k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh6 15.Kf6 Kh7 16.Nb5 Kh8 17.Bh1 Kg8 18.Nc7 Kh7 19.Bf3 Kh6 20.Be4 Kh5 21.Nb5 Kg4 22.Ke5
42/44 00:01 9,022k 5,728k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh6 15.Nd5 Kg5 16.Be4 Kh5 17.Nc7 Kh6 18.Kf6 Kh5 19.Bh1 Kh4 20.Ke5 Kh5 21.Nd5 Kh6 22.Kf6 Kh7
43/46 00:01 10,428k 5,727k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh6 15.Nd5 Kg7 16.Be4 Kh8 17.Ke6 Kg7 18.Nc7 Kf8 19.Kf6 Kg8 20.Nd5 Kh8 21.Ke5 Kg7 22.Nc7 Kg8 23.Bh1 Kf7
44/48 00:02 11,745k 5,752k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh7 15.Kf6 Kg8 16.Bd5+ Kh7 17.Be4+ Kh8 18.Bg2 Kh7 19.Bd5 Kh6 20.Be4 Kh5 21.Bf5 Kh6 22.Nd5 Kh5 23.Ne3
45/49 00:02 13,085k 5,844k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh7 15.Be4+ Kg8 16.Bh1 Kf7 17.Bd5+ Kg7 18.Na8 Kf8 19.Kf6 Ke8 20.Nc7+ Kd8 21.Nb5 Ke8 22.Be4 Kd8 23.Bh1 Kd7 24.Kf7 Kd8 25.Ke6 Ke8
46/49 00:02 14,934k 5,978k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kg8 15.Be4 Kg7 16.Ke6 Kh8 17.Kf6 Kg8 18.Nb5 Kh8 19.Bh1 Kg8 20.Nc7 Kh8 21.Nd5 Kh7 22.Be4+ Kg8 23.Ne7+ Kf8 24.Bh1 Ke8 25.Nd5
47/50 00:02 17,659k 6,017k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kg8 15.Bd5+ Kh7 16.Be4+ Kg7 17.Nb5 Kh8 18.Kf6 Kg8 19.Bd5+ Kh7 20.Nc7 Kh8 21.Bf3 Kh7 22.Be4+ Kg8 23.Nb5 Kh8 24.Bh1 Kg8 25.Bf3 Kh8
48/54 00:03 19,505k 6,073k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke3 Kh6 9.Ne8 Kg5 10.Kf3 Kh6 11.Kf4 Kh5 12.Bf3+ Kg6 13.Nc7 Kg7 14.Ke5 Kh6 15.Kf6 Kh7 16.Nd5 Kh8 17.Ne7 Kh7 18.Be4+ Kh8 19.Ng6+ Kh7 20.Nf4+ Kh8 21.Bd5 Kh7 22.Ne2 Kh8 23.Nc3 Kh7 24.Nb5 Kh6 25.Be4 Kh5 26.Nc7 Kg4 27.Ke5 Kh4
49/60 00:03 22,477k 6,182k +42.75 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg6 7.Be4+ Kg7 8.Ke5 Kg8 9.Bd5+ Kh7 10.Kf6 Kh8 11.Na8 Kh7 12.Nb6 Kh8 13.Nd7 Kh7 14.Be4+ Kg8 15.Ne5 Kh8 16.Ng6+ Kh7 17.Nf4+ Kh8 18.Bd5 Kh7 19.Ne2 Kh8 20.Nc3 Kh7 21.Nb5 Kh6 22.Be4 Kh5 23.Nc7 Kg4 24.Bh1 Kf4 25.Nd5+ Kg4 26.Be4 Kh5 27.Bf3+ Kh6 28.Ke5 Kg6 29.Nc7 Kh6 30.Nb5 Kg7
50/59+ 00:05 33,531k 6,293k +42.83 1.Kb2
50/59 00:07 46,858k 6,386k +42.83 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke5 Kh6 9.Ne6 Kh5 10.Nf4+ Kh6 11.Kf5 Kg7 12.Bd5 Kf8 13.Ng6+ Ke8 14.Ke6 Kd8 15.Kd6 Ke8 16.Bb3 Kd8 17.Ba4 Kc8 18.Nf8 Kb7 19.Nd7 Kc8 20.Bc6 Kd8 21.Nb8 Kc8 22.Na6 Kd8 23.Nc7 Kc8 24.Nb5 Kb8 25.Be4 Kc8
51/55 00:09 61,845k 6,502k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke5 Kh6 9.Kf6 Kh5 10.Bf5 Kh4 11.Ne6 Kh5 12.Nd8 Kh4 13.Ke5 Kg3 14.Ne6 Kf3 15.Kd4 Ke2 16.Nc7 Kd2 17.Be4 Kc1 18.Kc3 Kd1 19.Bf5 Ke1 20.Ne6 Ke2 21.Bg6 Kf3 22.Kd3 Kg4 23.Ke4 Kh4 24.Kf4 Kh3 25.Ke5 Kh4 26.Ke4 Kg3 27.Bh5
52/54 00:12 80,441k 6,627k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke5 Kh6 9.Kf6 Kh5 10.Bf5 Kh4 11.Ne6 Kh5 12.Nd8 Kh4 13.Ke5 Kg3 14.Ne6 Kf3 15.Kd4 Ke2 16.Nc7 Kd2 17.Be4 Kc1 18.Kc3 Kd1 19.Bf5 Ke1 20.Ne6 Ke2 21.Bg6 Kf3 22.Kd3 Kg4 23.Ke4 Kg3 24.Bf5 Kh4 25.Kd5 Kh5 26.Kd4 Kh6 27.Be4 Kh5
53/56 00:13 87,364k 6,670k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Ke5 Kh6 9.Kf6 Kh5 10.Bf5 Kh4 11.Ne6 Kh5 12.Nd8 Kh4 13.Ke5 Kg3 14.Ne6 Kf3 15.Kd4 Ke2 16.Nc7 Kd2 17.Nb5 Kc1 18.Kc3 Kd1 19.Nc7 Ke1 20.Ne6 Ke2 21.Bd3+ Ke3 22.Bg6 Kf3 23.Kb2 Ke3 24.Bf5 Kd2 25.Bb1 Ke3 26.Kc3 Ke2 27.Bf5 Ke1 28.Kd3 Kd1
54/58 00:14 96,602k 6,727k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Nd5 Kh4 9.Ke5 Kh5 10.Ne3 Kg5 11.Bf3 Kg6 12.Bd5 Kg7 13.Kd4 Kf6 14.Ba8 Ke7 15.Kc3 Kd8 16.Bh1 Ke7 17.Kd4 Kd6 18.Nd5 Kc6 19.Nc3+ Kd6 20.Nb5+ Ke6 21.Be4 Ke7 22.Ke5 Kd7 23.Ba8 Ke8 24.Nc7+ Kf7 25.Nd5 Ke8 26.Kd4 Kd7 27.Kd3 Kd6 28.Kc3
55/58 00:14 100,205k 6,738k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Nd5 Kh4 9.Ke5 Kh5 10.Ne3 Kg5 11.Bf3 Kg6 12.Bd5 Kg7 13.Bh1 Kh6 14.Kf6 Kh7 15.Be4+ Kg8 16.Ba8 Kh7 17.Bd5 Kh8 18.Bc6 Kh7 19.Ke5 Kh8 20.Bh1 Kg7 21.Nd5 Kh7 22.Be4+ Kg7 23.Ne3 Kf7 24.Kd4 Kg8 25.Ba8 Kh8 26.Ke5 Kg7 27.Nd5 Kf7 28.Nc7 Kg7 29.Bd5
56/59 00:16 109,571k 6,785k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Nd5 Kh6 9.Nc3 Kh5 10.Nb5 Kh6 11.Ke5 Kg7 12.Ba8 Kf7 13.Nc7 Kg7 14.Bd5 Kh7 15.Be4+ Kg7 16.Ne6+ Kh8 17.Bc6 Kg8 18.Kf6 Kh7 19.Bf3 Kh6 20.Nc7 Kh7 21.Ke5 Kh6 22.Na8 Kh7 23.Be4+ Kh6 24.Bd5 Kh7 25.Nc7 Kg7 26.Ne6+ Kh7 27.Bh1 Kh6 28.Be4 Kh5 29.Nf4+ Kh6 30.Kf5
57/63 00:17 116,227k 6,826k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg4 7.Be4 Kg5 8.Nd5 Kh6 9.Nc3 Kh5 10.Nb5 Kh6 11.Ba8 Kg7 12.Bd5 Kh7 13.Nc7 Kg7 14.Ke5 Kh7 15.Bf3 Kh6 16.Kf6 Kh7 17.Be4+ Kh8 18.Nd5 Kg8 19.Nc3 Kh8 20.Nb5 Kg8 21.Bd5+ Kh8 22.Nc7 Kh7 23.Ke5 Kg7 24.Kf5 Kf8 25.Bh1 Kg7 26.Nd5 Kh6 27.Kf6 Kh7 28.Ne3 Kg8 29.Bd5+ Kh7 30.Ke5 Kh6 31.Be4 Kg7 32.Ba8
58/64 00:19 132,258k 6,886k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg6 7.Be4+ Kf7 8.Bh1 Kg8 9.Ke4 Kf7 10.Ke5 Kg6 11.Ba8 Kg7 12.Bd5 Kh7 13.Kf6 Kh8 14.Kf5 Kg7 15.Bh1 Kh6 16.Kf6 Kh7 17.Nd5 Kh8 18.Ne3 Kg8 19.Bd5+ Kh7 20.Kf5 Kg7 21.Ke5 Kh6 22.Nc4 Kg5 23.Bc6 Kg6 24.Be4+ Kg7 25.Ne3 Kf7 26.Bd5+ Kg7 27.Nc2 Kh8 28.Kf6 Kh7 29.Ba8 Kh6 30.Ne3 Kh7 31.Bd5 Kh8
59/66 00:19 137,521k 6,889k +42.87 1.Kb2 Kc4 2.Kc2 Kd4 3.Nc7 Ke5 4.Kc3 Kf6 5.Kd4 Kf5 6.Bd5 Kg5 7.Ke5 Kh6 8.Bf3 Kh7 9.Be4+ Kg7 10.Nb5 Kh8 11.Bh1 Kg7 12.Nc7 Kh6 13.Kd4 Kg6 14.Be4+ Kf7 15.Ba8 Kg7 16.Ke4 Kh8 17.Ke5 Kg7 18.Nd5 Kh6 19.Ne3 Kg7 20.Nc2 Kg8 21.Bd5+ Kg7 22.Ne3 Kh7 23.Kf6 Kh8 24.Be4 Kg8 25.Nf5 Kh7 26.Ne7+ Kh8 27.Ba8 Kh7 28.Bd5 Kh8 29.Nc6 Kh7 30.Na5 Kh6 31.Nb3 Kh7
60/64+ 00:28 201,441k 6,983k +42.95 1.Bf3
60/64+ 01:20 542,090k 6,735k +43.02 1.Bf3
60/66+ 01:40 669,816k 6,693k +43.13 1.Bf3
60/66+ 02:13 888,219k 6,634k +43.29 1.Bf3
60/66+ 03:07 1,232,448k 6,580k +43.51 1.Bf3
60/66+ 03:55 1,531,476k 6,502k +43.80 1.Bf3
60/70+ 04:30 1,740,582k 6,434k +44.18 1.Bf3
60/71+ 04:48 1,847,468k 6,411k +44.68 1.Bf3
60/71+ 05:15 2,008,110k 6,373k +45.33 1.Bf3
60/71+ 05:50 2,206,469k 6,303k +46.15 1.Bf3
60/71+ 06:13 2,333,858k 6,255k +47.20 1.Bf3
60/73+ 06:58 2,585,183k 6,172k +48.54 1.Bf3
60/73+ 07:22 2,715,896k 6,133k +50.22 1.Bf3
60/73+ 07:44 2,837,360k 6,104k +52.35 1.Bf3
60/73+ 08:04 2,942,029k 6,075k +55.03 1.Bf3
60/73+ 08:16 3,009,805k 6,062k +58.40 1.Bf3
60/80+ 08:27 3,069,103k 6,051k +62.63 1.Bf3
60/81+ 08:37 3,123,459k 6,038k +67.93 1.Bf3
60/81+ 08:49 3,190,596k 6,023k +74.58 1.Bf3
60/82+ 08:59 3,247,389k 6,020k +82.92 1.Bf3
60/82+ 09:08 3,299,636k 6,013k +93.35 1.Bf3
60/82+ 09:15 3,341,387k 6,011k +106.42 1.Bf3
60/82+ 09:20 3,366,975k 6,007k +122.77 1.Bf3

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 3:29 pm
by Uri Blass
Here is the end of analysis when stockfish see the mate by search.

[D]N7/8/8/8/3k4/8/8/K6B w - - 0 1

Stockfish_17101111_x64_modern:

60/82+ 09:20 3,366,975k 6,007k +122.77 1.Bf3
60/93 28:53 9,819,484k 5,666k +M33 1.Nc7 Ke5 2.Ne8 Kf4 3.Kb2 Kf5 4.Bc6 Kf4 5.Kc3 Ke5 6.Kc4 Kf4 7.Kd4 Kf5 8.Bd5 Kg6 9.Ke5 Kh7 10.Be4+ Kh8 11.Bc2 Kg8 12.Kf6 Kh8 13.Nd6 Kg8 14.Nf7 Kf8 15.Bh7 Ke8 16.Ne5 Kd8 17.Be4 Ke8 18.Bd5 Kf8 19.Nd7+ Ke8 20.Bc6 Kd8 21.Ne5 Kc8 22.Ke7 Kc7 23.Bd7 Kb6 24.Kd6 Ka5 25.Kc5 Ka6 26.Nc4 Kb7 27.Kd6 Kb8 28.Kc6 Ka7 29.Bc8 Kb8 30.Nd6 Ka7 31.Kc7 Ka8 32.Bb7+ Ka7 33.Nb5+

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 5:59 pm
by Rodolfo Leoni
Uri Blass wrote:
I believe that after enough time alphazero will learn.
In the same way that it learned about the opening position it can learn every new position.

....................................................................
Crazy idea: it'd be funny to check if AlphaZ can just go through TBs files to learn from them.

Not so crazy, maybe.

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 7:47 pm
by syzygy
Rodolfo Leoni wrote:
Uri Blass wrote:
I believe that after enough time alphazero will learn.
In the same way that it learned about the opening position it can learn every new position.

....................................................................
Crazy idea: it'd be funny to check if AlphaZ can just go through TBs files to learn from them.

Not so crazy, maybe.
It will learn any position you give it, but you'll have to accept it will forget many others. Neural networks can only hold so much information and do so much computation. A well-compressed TB is vastly more efficient at this.

AlphaZero is not a miracle that defeats laws of nature (and laws of computation and information).

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 8:11 pm
by corres
[quote="Rodolfo Leoni"]
AlphaZ learning couldn't compete with Romi learning on traditional hardware.
[/quote]

AlphaZero can not work on any traditional hardware and never will be a Windows version of AlphaZero.
But you are right, if somebody would give Stockfis the Romi learning system this derivative of Stockfish would be stronger - at least in a kind of aspect.

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Tue Dec 12, 2017 8:52 pm
by corres
[quote="Rodolfo Leoni"]
I think it's not matter of AlphaZero. AlphaZero is only the first raw software on a new spectacular hardware system. As I think all best computer chess programmers are here, I wonder what wonderful engines you, Marco, Mark Lefler, and any top one could build with that machine.
[/quote]

Hardware of AlphaZero and its software are belong together. They are part of a very complete system.
Sorry, but I have hard doubt about there is anybody between the well-known chess programmer who can write the programs(!) of AlphaZero. It is not a chance that a big team worked on it.

[quote="Rodolfo Leoni"]
And I think at a graph: 20 years ago there were several hundred thousends correspondence chess players. As engines strenght grew the number of players diminished. They are about 5000 at ICCF, at the present moment.
The % of drawish lines grew too. Now, with an hardware like that (and much stronger engines), I guess 99% of games could end as a draw. It's like an equalizer. On correspondence chess TCs, that equalizer could really vanquish the pleasure of playing a game.
Maybe it's a bit too pessimistic. The hardware isn't available, yet, and there are lots of things which one cannot understand about it. Let us see. [/quote]

I agree you.
You are not pessimistic. It is pity but this is the reality, I think so.
I am afraid the era of the Personal Computer and the personal software
come to their end. We are at the beginning of the new era: the era of the big networks - this gives the base of AlphaZero too.

Re: AlphaGo Zero And AlphaZero, RomiChess done better

Posted: Wed Dec 13, 2017 8:45 am
by Rodolfo Leoni
syzygy wrote:
Rodolfo Leoni wrote:
Uri Blass wrote:
I believe that after enough time alphazero will learn.
In the same way that it learned about the opening position it can learn every new position.

....................................................................
Crazy idea: it'd be funny to check if AlphaZ can just go through TBs files to learn from them.

Not so crazy, maybe.
It will learn any position you give it, but you'll have to accept it will forget many others. Neural networks can only hold so much information and do so much computation. A well-compressed TB is vastly more efficient at this.

AlphaZero is not a miracle that defeats laws of nature (and laws of computation and information).
All of that sounds quite interesting. So you say it can't "remember" the zillion 7 men TBs root positions, but all the "knowledge" it gets goes minimaxed up to startposition. And it can't keep traditional data, but a human-like understanding of what it propagated.

No real hash tables, IMO. But if/when hardware will be available a new TB structure could be done, maybe. (?)