How long do we need to wait for the next trained 40x512 net?
You should retest. I already tested both 40x512 nets. And they find it quick for 1 second, and change their minds. And it took a long time again to see Rxe6.
Just look at the eval he has.
Yes don't goat the test.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
How long do we need to wait for the next trained 40x512 net?
You should retest. I already tested both 40x512 nets. And they find it quick for 1 second, and change their minds. And it took a long time again to see Rxe6.
Just look at the eval he has.
The question was not:
What is the correct evaluation?
The question was also not:
What is the fastest mate?
The question was:
How long does it take for your NN engine to find Rxe6?
And 40x512 is the fastest one at finding Rxe6.
Of course the net is not perfectly yet.
It needs much more training.
How long do we need to wait for the next trained 40x512 net?
You should retest. I already tested both 40x512 nets. And they find it quick for 1 second, and change their minds. And it took a long time again to see Rxe6.
Just look at the eval he has.
The question was not:
What is the correct evaluation?
The question was also not:
What is the fastest mate?
The question was:
How long does it take for your NN engine to find Rxe6?
And 40x512 is the fastest one at finding Rxe6.
Of course the net is not perfectly yet.
It needs much more training.
You got me!
Now how long to see a mate or big winning score....
In that case it is not even close to being the fastest net.
40x512 net after 3 minuets. The real results.
New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
LS 13.2 pops it up from -200 eval to finding the right line with very good eval at just over 275k nodes, and gets to about cp 3000 at 500k nodes with b1b2 in 2nd place at -9 cp.
jjoshua2 wrote: ↑Sat Feb 08, 2020 5:38 pm
LS 13.2 pops it up from -200 eval to finding the right line with very good eval at just over 275k nodes, and gets to about cp 3000 at 500k nodes with b1b2 in 2nd place at -9 cp.
I was able to do the same with Lc0 + any NN. And was able to cut the time down to as low as 14 secs. By tuning for this position. Showing this is a search issue with Lc0, not a NN issue.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
jjoshua2 wrote: ↑Sat Feb 08, 2020 5:38 pm
LS 13.2 pops it up from -200 eval to finding the right line with very good eval at just over 275k nodes, and gets to about cp 3000 at 500k nodes with b1b2 in 2nd place at -9 cp.
I was able to do the same with Lc0 + any NN. And was able to cut the time down to as low as 14 secs. By tuning for this position. Showing this is a search issue with Lc0, not a NN issue.
I don't agree. You can tune Houdini, SF, Rybka, and others just as easily for specific situations, but that does not mean their default settings are at fault. The settings are supposed to be tuned for optimal overall play, so cranking up the tactical sensitivity will almost certainly come at a cost. Just put the CPUCT at 5.0 or something even more absurd, and no doubt the solution will come extremely fast, but the overall Elo will drop significantly. Still, I wouldn't read too much in one position here or there. Computer chess veterans know all too well there is always some engine that will find Position X in record speed, but that doesn't mean it is the best engine.
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
jjoshua2 wrote: ↑Sat Feb 08, 2020 5:38 pm
LS 13.2 pops it up from -200 eval to finding the right line with very good eval at just over 275k nodes, and gets to about cp 3000 at 500k nodes with b1b2 in 2nd place at -9 cp.
I was able to do the same with Lc0 + any NN. And was able to cut the time down to as low as 14 secs. By tuning for this position. Showing this is a search issue with Lc0, not a NN issue.
I don't agree. You can tune Houdini, SF, Rybka, and others just as easily for specific situations, but that does not mean their default settings are at fault. The settings are supposed to be tuned for optimal overall play, so cranking up the tactical sensitivity will almost certainly come at a cost. Just put the CPUCT at 5.0 or something even more absurd, and no doubt the solution will come extremely fast, but the overall Elo will drop significantly. Still, I wouldn't read too much in one position here or there. Computer chess veterans know all too well there is always some engine that will find Position X in record speed, but that doesn't mean it is the best engine.
I agree. I am just suggesting that Lc0 can improve the search element of the NN. And it is not the NN itself that is to blame. Lc0 search is still young, and needs refinement. A/B search took years to develop.
I agree, and did not tune CPUCT. I tune by policy temp and FPU value.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Really nice, especially that it comes from a subvariation of a very well known human game and I am sure Kasparov would find 33.Rxe6 quickly.
One more time it looks like bug in search. I've already posted about it a few times on both Discord and Lc0 forum. There are positions where it sometimes find the solution in 3 seconds and sometimes doesn't find it even after very long time (like 10 minutes). If you run it several times with the newest T60 you will see that behavior: it's either there in like 5-10 seconds or never gets there.
Maybe it's a GUI issue or god knows what. I am running it in SCIDvsPC and usually the first run doesn't find it but then after killing the engine and restarting it it finds it instantly with +97% score. You can confirm in the task manager that it indeed kills Lc0 process in the meantime. Something is terribly wrong with the search implementation there.
It's similar with the famous Bg5 move of Alpha Zero. It sometimes finds it quickly and usually doesn't find it even after 30 minutes on 2080ti.
I've signaled it in the past and it seems to me it's not considered a serious problem by the devs. From my perspective it means running Lc0 with more than 1 thread just doesn't work and users who use it for analysis or position testing should be warned about random blunders occurring with more than one thread. I would never run any kind of analysis with 2+ threads if I knew about the extent of this behavior.
OneTrickPony wrote: ↑Sun Feb 09, 2020 10:40 am
Really nice, especially that it comes from a subvariation of a very well known human game and I am sure Kasparov would find 33.Rxe6 quickly.
One more time it looks like bug in search. I've already posted about it a few times on both Discord and Lc0 forum. There are positions where it sometimes find the solution in 3 seconds and sometimes doesn't find it even after very long time (like 10 minutes). If you run it several times with the newest T60 you will see that behavior: it's either there in like 5-10 seconds or never gets there.
Maybe it's a GUI issue or god knows what. I am running it in SCIDvsPC and usually the first run doesn't find it but then after killing the engine and restarting it it finds it instantly with +97% score. You can confirm in the task manager that it indeed kills Lc0 process in the meantime. Something is terribly wrong with the search implementation there.
It's similar with the famous Bg5 move of Alpha Zero. It sometimes finds it quickly and usually doesn't find it even after 30 minutes on 2080ti.
I've never seen what you describe. You should run Lc0 from a command line, to rule out GUI-related issues.
Here are five runs of the position in this thread, with Lc0 terminated after each run. No GUI was used. Default settings, except for Backend = cudnn-fp16.
You got me!