EPR = Early Ply Reductions and seems to be (way?) stronger than LMR when tested totally separate. However, they should be able to be combined for even greater effect.
The variance that I am testing now is:
If depth > iDepth / 2 reduce search for all moves by 3 instead of by 1 and research at a reduction of 1 if the score returned is > alpha. It is just a simple start and testing is still early, but so far it looks impressive!
EPR, even better than LMR!
Moderators: hgm, Rebel, chrisw
-
- Posts: 3196
- Joined: Fri May 26, 2006 3:00 am
- Location: WY, USA
- Full name: Michael Sherwin
EPR, even better than LMR!
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
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
-
- Posts: 327
- Joined: Sat Mar 27, 2010 7:15 pm
Re: EPR, even better than LMR!
interesting.. what is iDepth?Michael Sherwin wrote:EPR = Early Ply Reductions and seems to be (way?) stronger than LMR when tested totally separate. However, they should be able to be combined for even greater effect.
The variance that I am testing now is:
If depth > iDepth / 2 reduce search for all moves by 3 instead of by 1 and research at a reduction of 1 if the score returned is > alpha. It is just a simple start and testing is still early, but so far it looks impressive!
can you post some test results?
-
- Posts: 3196
- Joined: Fri May 26, 2006 3:00 am
- Location: WY, USA
- Full name: Michael Sherwin
Re: EPR, even better than LMR!
iDepth is the initial depth of the searchsilentshark wrote:interesting.. what is iDepth?Michael Sherwin wrote:EPR = Early Ply Reductions and seems to be (way?) stronger than LMR when tested totally separate. However, they should be able to be combined for even greater effect.
The variance that I am testing now is:
If depth > iDepth / 2 reduce search for all moves by 3 instead of by 1 and research at a reduction of 1 if the score returned is > alpha. It is just a simple start and testing is still early, but so far it looks impressive!
can you post some test results?
My great test results have turned ordinary with the addition of a couple hundred more games. If I can improve the idea then I will post some results. I just got too excited after only 60+ games. I should have known better.
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
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
-
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: EPR, even better than LMR!
Don't worry, I have done the same thing many times - get real excited over a great start then only to suffer disappointment.Michael Sherwin wrote:iDepth is the initial depth of the searchsilentshark wrote:interesting.. what is iDepth?Michael Sherwin wrote:EPR = Early Ply Reductions and seems to be (way?) stronger than LMR when tested totally separate. However, they should be able to be combined for even greater effect.
The variance that I am testing now is:
If depth > iDepth / 2 reduce search for all moves by 3 instead of by 1 and research at a reduction of 1 if the score returned is > alpha. It is just a simple start and testing is still early, but so far it looks impressive!
can you post some test results?
My great test results have turned ordinary with the addition of a couple hundred more games. If I can improve the idea then I will post some results. I just got too excited after only 60+ games. I should have known better.
I have been all over the ground you are trying to cover with this. I think we have wasted more time fooling around with LMR than any other single aspect of the program.
-
- Posts: 3196
- Joined: Fri May 26, 2006 3:00 am
- Location: WY, USA
- Full name: Michael Sherwin
Re: EPR, even better than LMR!
Thanks for the shoulder!
This was the most exciting start ever and therefore the biggest let down.
LMR is good for just so much--it probably is a waste of valuable time to try to get more out of it. I should be happy that I was able to get a bigger elo jump with a LMR reduction of 2 ply instead of just 1.
Null move is extremely efficient at eliminating any position that 'must' fall outside the search bounds. It must be more than coincidence that Olithink, RomiChess, Stockfish and maybe others all settled upon the exact same depth reduction schedule, AFAIK. It was a good thing that a little more was squeezed from null move, now it is time to look elsewhere.
Extensions other than the standard extension for check seem to be a waste of time also. Maybe single reply extensions are helpful but there is conflicting info on that.
It seems that there has been progress lately in move ordering techniques. How else would the newer engines be searching so deep compared to just awhile ago. Is it all due to just ordering at the root based on node count? There must be more, but what is it. A well tunned eval will help move ordering--is that what's doing it?
YES, as you imply, I need to change my focus if my program is to progress any further!
This was the most exciting start ever and therefore the biggest let down.
LMR is good for just so much--it probably is a waste of valuable time to try to get more out of it. I should be happy that I was able to get a bigger elo jump with a LMR reduction of 2 ply instead of just 1.
Null move is extremely efficient at eliminating any position that 'must' fall outside the search bounds. It must be more than coincidence that Olithink, RomiChess, Stockfish and maybe others all settled upon the exact same depth reduction schedule, AFAIK. It was a good thing that a little more was squeezed from null move, now it is time to look elsewhere.
Extensions other than the standard extension for check seem to be a waste of time also. Maybe single reply extensions are helpful but there is conflicting info on that.
It seems that there has been progress lately in move ordering techniques. How else would the newer engines be searching so deep compared to just awhile ago. Is it all due to just ordering at the root based on node count? There must be more, but what is it. A well tunned eval will help move ordering--is that what's doing it?
YES, as you imply, I need to change my focus if my program is to progress any further!
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
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
-
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: EPR, even better than LMR!
Move ordering at the root is just a very minor thing - it's my belief that LMR is heavily dependent on good move ordering everywhere - you can cheat more if you have good move ordering.Michael Sherwin wrote:Thanks for the shoulder!
This was the most exciting start ever and therefore the biggest let down.
LMR is good for just so much--it probably is a waste of valuable time to try to get more out of it. I should be happy that I was able to get a bigger elo jump with a LMR reduction of 2 ply instead of just 1.
Null move is extremely efficient at eliminating any position that 'must' fall outside the search bounds. It must be more than coincidence that Olithink, RomiChess, Stockfish and maybe others all settled upon the exact same depth reduction schedule, AFAIK. It was a good thing that a little more was squeezed from null move, now it is time to look elsewhere.
Extensions other than the standard extension for check seem to be a waste of time also. Maybe single reply extensions are helpful but there is conflicting info on that.
It seems that there has been progress lately in move ordering techniques. How else would the newer engines be searching so deep compared to just awhile ago. Is it all due to just ordering at the root based on node count? There must be more, but what is it. A well tunned eval will help move ordering--is that what's doing it?
YES, as you imply, I need to change my focus if my program is to progress any further!
-
- Posts: 4185
- Joined: Tue Mar 14, 2006 11:34 am
- Location: Ethiopia
Re: EPR, even better than LMR!
No no micheal, squeeze harderLMR is good for just so much--it probably is a waste of valuable time to try to get more out of it. I should be happy that I was able to get a bigger elo jump with a LMR reduction of 2 ply instead of just 1.
I used reductions of 2 in 2006 IIRC, nowadays big reductions of 4 plies in both LMR and Null move are used in the strongest engines. Also futility pruning at shallower (8 plies ?) is used on top of that. The 101 of chess programming is to equal search depths of your opponents before you work on eval.
-
- Posts: 1154
- Joined: Fri Jun 23, 2006 5:18 am
Re: EPR, even better than LMR!
I believe there is much more to be gained from pruning, extensions, and reductions than has been gained so far (by any program). But as Don said, the dependencies (move ordering, history, evaluation/eval approximations, whatever) effect how much you can gain. Thus, if you feel you have reached a local maxima, it may be good to work on these other areas, and eventually come back.
-Sam
-Sam
-
- Posts: 454
- Joined: Sat Apr 04, 2009 6:44 pm
- Location: Bulgaria
Re: EPR, even better than LMR!
I had the same experience. Did so many tests to jump over the 1 ply reduction with no success. The negative impact of the dependencies still hits my attempts, so I know perfectly how you feel. When I get tired from it, just switching to another subject, then get back to it again, this is my "refreshing" approach .
-
- Posts: 3196
- Joined: Fri May 26, 2006 3:00 am
- Location: WY, USA
- Full name: Michael Sherwin
Re: EPR, even better than LMR!
So, I reduce like crazy to better other engines in search depth while loosing several hundred elo. Then the eval and move ordering is improved until all the elo is regained and much more. Sounds like a lesson for playing the stock market--average down by using margin and hope a bottom is reached before you get a margin call. I'm selling off 'good' stocks to average down on BP. I wonder if it is close to a bottom. I do have some margin left to chase it further, but I am hesitant to use it.Daniel Shawul wrote:No no micheal, squeeze harderLMR is good for just so much--it probably is a waste of valuable time to try to get more out of it. I should be happy that I was able to get a bigger elo jump with a LMR reduction of 2 ply instead of just 1.
I used reductions of 2 in 2006 IIRC, nowadays big reductions of 4 plies in both LMR and Null move are used in the strongest engines. Also futility pruning at shallower (8 plies ?) is used on top of that. The 101 of chess programming is to equal search depths of your opponents before you work on eval.
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
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