Dann Corbit wrote: ↑Tue Oct 29, 2019 5:51 pm
Since LC0 performed so terribly (I gave 12 minutes per position on a test set meant to run for 15 seconds per position) and is beaten by all the strong tactical engines by a landslide, I can only conclude that tactics are not very important for the game of chess.
LC0 has only a few-hundred thousand nodes per second. This means it is significantly stronger at tactics than any human player, but still far slower than the millions or even hundred-millions nodes-per-second of classic A/B engines.
Tactics remain important: but LC0 has "enough" tactics to play decently. Where it wins is that LC0 has very strong positional play / long-term skill in chess.
Dann Corbit wrote: ↑Tue Oct 29, 2019 5:51 pm
I did an experiment (perhaps flawed because I used this net:128x10-2019_1026_0142_45_230.pb) and LC0 is really terrible at tactics on a Nvidia TI-1080.
The engine crashed once when I was analyzing, so I restarted at the point where analysis left off, and it did not crash a second time.
I am re-running the test with a 256 net 256x20-2019_0726_0906_48_558.pb
Since LC0 performed so terribly (I gave 12 minutes per position on a test set meant to run for 15 seconds per position) and is beaten by all the strong tactical engines by a landslide, I can only conclude that tactics are not very important for the game of chess.
On the other hand, if they are important, and they can somehow be compiled into an LC0 net, perhaps LC0 would become utterly unbeatable
good afternoon, good morning and good night everyone; Let me differ from the sentence that "tactics are not important in chess"; In the year of 1977 I started to play chess in my 17 years as a self-taught; unfortunately I have never belonged to any chess school nor have I ever had a chess teacher to teach; My first bibliographical teachings were the book of the commented games of Fischer. At that time the teaching came directly from the paper works, nothing electronic. I always dreamed and wanted to train chess with a teacher and be able to play with this or this chess personally on a wooden chess table and board. But I could never do it; It's quite sad but not it. Such an orphan led me to take refuge and face the chess machines alone since 1980. I never received advice on how and how to train to be strong in chess. so many years passed without training methodology. The positive part of all this is to have allowed to develop the own criteria about the game of chess. Only after thousands of hours of practice and experience, draw and conclude that the most important thing in the chess player is the tactic. It is precisely the point that makes it more or less bright. So if someone asks me what advice to receive in chess training is: “Practice chess problems daily; and if you want to be a strong person, dedicate time to sport, chess and prayer, so will your soul, your body and your mind, be strong, through integral gymnastics. ”
Dann Corbit wrote: ↑Tue Oct 29, 2019 5:51 pm
Since LC0 performed so terribly (I gave 12 minutes per position on a test set meant to run for 15 seconds per position) and is beaten by all the strong tactical engines by a landslide, I can only conclude that tactics are not very important for the game of chess.
LC0 has only a few-hundred thousand nodes per second. This means it is significantly stronger at tactics than any human player, but still far slower than the millions or even hundred-millions nodes-per-second of classic A/B engines.
Tactics remain important: but LC0 has "enough" tactics to play decently. Where it wins is that LC0 has very strong positional play / long-term skill in chess.
I suspect that I made a bad net choice, so I am replaying.
But I gave LC0 48 times more time per position, and the results were still terrible.
So I suspect your diagnosis is correct.
I also expect the bigger net to do better
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Dann Corbit wrote: ↑Tue Oct 29, 2019 5:51 pm
Since LC0 performed so terribly (I gave 12 minutes per position on a test set meant to run for 15 seconds per position) and is beaten by all the strong tactical engines by a landslide, I can only conclude that tactics are not very important for the game of chess.
LC0 has only a few-hundred thousand nodes per second. This means it is significantly stronger at tactics than any human player, but still far slower than the millions or even hundred-millions nodes-per-second of classic A/B engines.
Tactics remain important: but LC0 has "enough" tactics to play decently. Where it wins is that LC0 has very strong positional play / long-term skill in chess.
I suspect that I made a bad net choice, so I am replaying.
But I gave LC0 48 times more time per position, and the results were still terrible.
So I suspect your diagnosis is correct.
I also expect the bigger net to do better
Smaller nets probably will be better in tactics, while bigger nets will probably be stronger for positional play. A bigger net means spending more time evaluating the neural net... which necessarily means having fewer-and-fewer nodes-per-second.
Case in point: a small net may have 10-microseconds per node (100,000 nodes per second), while a big net may have 50 microseconds per node (20,000 nodes per second).
Last edited by dragontamer5788 on Tue Oct 29, 2019 8:56 pm, edited 1 time in total.
Another possibility is that the bigger nets become "more sagacious".
A super GM will often see a tactical move after a very brief study.
There are also the instant tactical genius people like Morphy and Tal, but I think that probably cannot happen with a net
You are probably right, but I am going to run the experiment anyway. Only way to know for sure that I can think of.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Dann Corbit wrote: ↑Tue Oct 29, 2019 8:55 pm
Another possibility is that the bigger nets become "more sagacious".
That's true. LC0's neural net evaluates not only the probability of win, but also which moves are more "interesting", and where the MCTS search should focus more of its time on.
Its certainly worth a test, but my expectation is that "highly tactical" positions are ones which require a deep alpha-beta search... one that "breaks" a number of heuristics and feels unnatural to play. As such, any testing on "only tactical" positions will probably favor the dumb brute-force engines.
Dann Corbit wrote: ↑Tue Oct 29, 2019 5:51 pm
Since LC0 performed so terribly (I gave 12 minutes per position on a test set meant to run for 15 seconds per position) and is beaten by all the strong tactical engines by a landslide, I can only conclude that tactics are not very important for the game of chess.
LC0 has only a few-hundred thousand nodes per second. This means it is significantly stronger at tactics than any human player, but still far slower than the millions or even hundred-millions nodes-per-second of classic A/B engines.
Tactics remain important: but LC0 has "enough" tactics to play decently. Where it wins is that LC0 has very strong positional play / long-term skill in chess.
I suspect that I made a bad net choice, so I am replaying.
But I gave LC0 48 times more time per position, and the results were still terrible.
So I suspect your diagnosis is correct.
I also expect the bigger net to do better
Smaller nets probably will be better in tactics, while bigger nets will probably be stronger for positional play. A bigger net means spending more time evaluating the neural net... which necessarily means having fewer-and-fewer nodes-per-second.
Case in point: a small net may have 10-microseconds per node (100,000 nodes per second), while a big net may have 50 microseconds per node (20,000 nodes per second).
Not necessarily, there are policy and value parts affecting the tactics, not just NPS. It depends on say time control too. I have mixed results on that on tactical suites like Arasan and trimmed WAC. Bigger nets seem to improve on tactics faster with longer TC than smaller nets. But many smaller nets start from better tactics at very short TC.
Dann Corbit wrote: ↑Tue Oct 29, 2019 5:51 pm
I did an experiment (perhaps flawed because I used this net:128x10-2019_1026_0142_45_230.pb) and LC0 is really terrible at tactics on a Nvidia TI-1080.
The engine crashed once when I was analyzing, so I restarted at the point where analysis left off, and it did not crash a second time.
I am re-running the test with a 256 net 256x20-2019_0726_0906_48_558.pb
Since LC0 performed so terribly (I gave 12 minutes per position on a test set meant to run for 15 seconds per position) and is beaten by all the strong tactical engines by a landslide, I can only conclude that tactics are not very important for the game of chess.
On the other hand, if they are important, and they can somehow be compiled into an LC0 net, perhaps LC0 would become utterly unbeatable
good afternoon, good morning and good night everyone; Let me differ from the sentence that "tactics are not important in chess"; In the year of 1977 I started to play chess in my 17 years as a self-taught; unfortunately I have never belonged to any chess school nor have I ever had a chess teacher to teach; My first bibliographical teachings were the book of the commented games of Fischer. At that time the teaching came directly from the paper works, nothing electronic. I always dreamed and wanted to train chess with a teacher and be able to play with this or this chess personally on a wooden chess table and board. But I could never do it; It's quite sad but not it. Such an orphan led me to take refuge and face the chess machines alone since 1980. I never received advice on how and how to train to be strong in chess. so many years passed without training methodology. The positive part of all this is to have allowed to develop the own criteria about the game of chess. Only after thousands of hours of practice and experience, draw and conclude that the most important thing in the chess player is the tactic. It is precisely the point that makes it more or less bright. So if someone asks me what advice to receive in chess training is: “Practice chess problems daily; and if you want to be a strong person, dedicate time to sport, chess and prayer, so will your soul, your body and your mind, be strong, through integral gymnastics. ”
Sincerely,
Pablo
ouch, what a thread abuse...
Táctical are really important.
Personally I prefer chess machines in a relaxion directly proportional to speed, tactical power and algorithm and I prefer the human player whom see further in tactical combinations. If chess were solved in a database, to my liking, chess would lose all interest and become obsolete chess.
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
Dann Corbit wrote: ↑Tue Oct 29, 2019 8:55 pm
You are probably right, but I am going to run the experiment anyway. Only way to know for sure that I can think of.
It is early, but 10/28 so far (running the second set first with the bigger net)
Analysis of G:\chess\epd\ACT2.epd
Analyzing engine: Lc0
10/29/2019 9:25:14 AM Level: 720 Seconds
Source : List of EPD/PGN files
G:\chess\epd\ACT2.epd
G:\chess\epd\ACT1-B.epd
Colours : White moves=True, Black moves=True
Direction : forward
Games (PGN) : From 1 to 999999
Moves : From 1 to 250
Positions (EPD) : From 1 to 999999
Use Engine(s) : List
Lc0
Use seconds per move : 720 Seconds per move
Analysis Lines : Minimum search depth=2
Activate abort analysis : False
1) Nd5-b4 a3-a4
2) Ba8-h1 Ba8-h1 * 10 Seconds
3) e6xd7 Bf1-a6
4) Rg1-h1 Rg1-h1 * 7 Seconds
5) Nb7-d8 Nb7-d8 * 2 Seconds
6) Qc8-c3 Be6xd5
7) Rf5-f8 g6xh7
8) Bb8xd6 Bb8xd6 * 166 Seconds
9) b2-b3 Nd5-f6
10) b6-b7 b6-b7 * 4 Seconds
11) Be2-c4 Be2-h5
12) Bc3-d2 Bc3-d2 * 4 Seconds
13) Kg7-h8 Ba7-d4
14) Ke6-f5 Ke6-f5 * 0 Seconds
15) b4-b5 b4-b5 * 1 Second
16) Rc2-c5 Qh8-h7
17) Kb3-c3 Nd8-c6
18) Bf6-g7 Kh3-h4
19) Qc1-a3 Qc1xc2
20) Ne5-c4 Ne5-f7
21) Rh4-a4 Rh4-a4 * 0 Seconds
22) Nb3-c1 Nf1-e3
23) Be5-b8 Be5-d4
24) .. Qd1-e1 Qd1-e1 * 6 Seconds
25) Rb1-a1 Rb1-d1
26) Ne1-g2 Ne1xd3
27) Qa1-f6 Nc1-b3
28) f3-f4 Ng7-e6
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.