However, having back-ported the major changes (read copied from your Github master) to Firenzina-2_3-xTreme everything runs fine.
Go figure.
In GitHub, there's a feature that shows to me (I am running Windows) if my local Firenzina is in sync with the master (or another branch, and we currently have just one other, if I select that one). Everything runs, that's good. But is your code in sync with the current master? Please take a look. This is a way to check if we have reached Linux portability yet.
Put simply - no. The code I have committed runs perfectly on Linux. I will sync with master, and try again.
I have managed to get cutechess to accept the compiles of Original and Aggressive - and after 1000 games each, the difference is well within error bars (Original is 1 Elo higher than Aggressive)
I think we can safely say there is no difference between the two.
ZirconiumX wrote:I have managed to get cutechess to accept the compiles of Original and Aggressive - and after 1000 games each, the difference is well within error bars (Original is 1 Elo higher than Aggressive)
I think we can safely say there is no difference between the two.
Matthew:out
This result is quite intriguing. I compared three Fire 2.2 compiles: INF DTU2, GH, and SK94z, all of which have settings different from Original. GH (settings a lot like Original) proved stronger than an earlier INF DTU2 (similar, but not identical to Conservative) and much stronger than SK94z (similar, but not identical to Aggressive). However, previous experience tells us that these differences may have more to do with the fact that different people (NS, GH, and SK) compiled the code using different compiler settings and, quite possibly, different compilers. Engine settings may not have played much of a role. I wonder if you can design a set of settings that we would call Ridiculous and test it to make sure that it plays poorly, visibly worse than Original, just so that we'd know that the user settings actually kick in and are not replaced by the defaults.
ZirconiumX wrote:I have managed to get cutechess to accept the compiles of Original and Aggressive - and after 1000 games each, the difference is well within error bars (Original is 1 Elo higher than Aggressive)
I think we can safely say there is no difference between the two.
Matthew:out
This result is quite intriguing. I compared three Fire 2.2 compiles: INF DTU2, GH, and SK94z, all of which have settings different from Original. GH (settings a lot like Original) proved stronger than an earlier INF DTU2 (similar, but not identical to Conservative) and much stronger than SK94z (similar, but not identical to Aggressive). However, previous experience tells us that these differences may have more to do with the fact that different people (NS, GH, and SK) compiled the code using different compiler settings and, quite possibly, different compilers. Engine settings may not have played much of a role. I wonder if you can design a set of settings that we would call Ridiculous and test it to make sure that it plays poorly, visibly worse than Original, just so that we'd know that the user settings actually kick in and are not replaced by the defaults.
I'm not going to say much, just this:
Robert Houdart (Houdini) ()
trying to use CLOP to improve Fire would probably require well over 100,000 games
Robert Houdart (Houdini) ()
it's not realistic to expect that it's possible to improve simultaneously 5 parameters in Fire 2.2 by playing 3000 games
Robert Houdart (Houdini) ()
Matthew, there is no low-hanging fruit in Fire / Ivanhoe
Robert Houdart (Houdini) ()
just tuning some parameters (even if done correctly, with massive numbers of games) will not bring you very far
He's definitely optimistic about Firenzina.
Roberto Munter
Add this to what Robert said: ..and this is not easy, and it takes a lot of hard work (and often much frustration!)
ZirconiumX wrote:I have managed to get cutechess to accept the compiles of Original and Aggressive - and after 1000 games each, the difference is well within error bars (Original is 1 Elo higher than Aggressive)
I think we can safely say there is no difference between the two.
Matthew:out
This result is quite intriguing. I compared three Fire 2.2 compiles: INF DTU2, GH, and SK94z, all of which have settings different from Original. GH (settings a lot like Original) proved stronger than an earlier INF DTU2 (similar, but not identical to Conservative) and much stronger than SK94z (similar, but not identical to Aggressive). However, previous experience tells us that these differences may have more to do with the fact that different people (NS, GH, and SK) compiled the code using different compiler settings and, quite possibly, different compilers. Engine settings may not have played much of a role. I wonder if you can design a set of settings that we would call Ridiculous and test it to make sure that it plays poorly, visibly worse than Original, just so that we'd know that the user settings actually kick in and are not replaced by the defaults.
I'm not going to say much, just this:
Robert Houdart (Houdini) ()
trying to use CLOP to improve Fire would probably require well over 100,000 games
Robert Houdart (Houdini) ()
it's not realistic to expect that it's possible to improve simultaneously 5 parameters in Fire 2.2 by playing 3000 games
Robert Houdart (Houdini) ()
Matthew, there is no low-hanging fruit in Fire / Ivanhoe
Robert Houdart (Houdini) ()
just tuning some parameters (even if done correctly, with massive numbers of games) will not bring you very far
He's definitely optimistic about Firenzina.
Roberto Munter
Add this to what Robert said: ..and this is not easy, and it takes a lot of hard work (and often much frustration!)
Matthew:out
Robertolito is giving a perfect example of a psychological war here....
Only he,the great one,can pull the rabbit out of the hat....
It's not the case regards,
Dr.D
_No one can hit as hard as life.But it ain’t about how hard you can hit.It’s about how hard you can get hit and keep moving forward.How much you can take and keep moving forward….
ZirconiumX wrote:I have managed to get cutechess to accept the compiles of Original and Aggressive - and after 1000 games each, the difference is well within error bars (Original is 1 Elo higher than Aggressive)
I think we can safely say there is no difference between the two.
Matthew:out
This result is quite intriguing. I compared three Fire 2.2 compiles: INF DTU2, GH, and SK94z, all of which have settings different from Original. GH (settings a lot like Original) proved stronger than an earlier INF DTU2 (similar, but not identical to Conservative) and much stronger than SK94z (similar, but not identical to Aggressive). However, previous experience tells us that these differences may have more to do with the fact that different people (NS, GH, and SK) compiled the code using different compiler settings and, quite possibly, different compilers. Engine settings may not have played much of a role. I wonder if you can design a set of settings that we would call Ridiculous and test it to make sure that it plays poorly, visibly worse than Original, just so that we'd know that the user settings actually kick in and are not replaced by the defaults.
I'm not going to say much, just this:
Robert Houdart (Houdini) ()
trying to use CLOP to improve Fire would probably require well over 100,000 games
Robert Houdart (Houdini) ()
it's not realistic to expect that it's possible to improve simultaneously 5 parameters in Fire 2.2 by playing 3000 games
Robert Houdart (Houdini) ()
Matthew, there is no low-hanging fruit in Fire / Ivanhoe
Robert Houdart (Houdini) ()
just tuning some parameters (even if done correctly, with massive numbers of games) will not bring you very far
He's definitely optimistic about Firenzina.
Roberto Munter
Add this to what Robert said: ..and this is not easy, and it takes a lot of hard work (and often much frustration!)
Matthew:out
I readily believe that hard work will be required to improve the engine.
Each game on my octal took about 5 seconds (40/1+0.05), and I can run 7 at once.
100,000 * 5 = 500,000 seconds / 7 = 71,428.6 seconds (call it 80,000 or so)
80,000 / 60 = 1333 minutes
1,333 / 60 = 22 hours
22 hours / 5.5 hours I can run it for = 4 continuous days.
I suggest we release 2.3 now, so we don't wait forever to get everything tuned up.
Is there anywhere else you can think of where we can add a small amount of Elo? One guy on TCEC mentioned a material imbalance table being an Elo loss - we'll have to see what we can do.
One area we can definitely improve on is the code readability, which is pretty poor.
However, having back-ported the major changes (read copied from your Github master) to Firenzina-2_3-xTreme everything runs fine.
Go figure.
In GitHub, there's a feature that shows to me (I am running Windows) if my local Firenzina is in sync with the master (or another branch, and we currently have just one other, if I select that one). Everything runs, that's good. But is your code in sync with the current master? Please take a look. This is a way to check if we have reached Linux portability yet.
Put simply - no. The code I have committed runs perfectly on Linux. I will sync with master, and try again.
Matthew:out
Matthew, Dimitry,
If you don't mind, can you please have these private discussions in private (comments on the github repo, email etc.)
I see you're trying hard to keep this thread on top of the first page... I don't think there's anything useful to anyone but you two in these discussions: IOW it's spamming.
I think everyone got the message: Firenzina is a Fire clone, and progress (for those interested) can be followed on GitHub.
Theory and practice sometimes clash. And when that happens, theory loses. Every single time.
Matew, ignore,
continue with the project,
Many if interested.
Many people go to these anticlones, doctors anything and
knows what all.
Always bothering with snares.
Matew, when I can enjoy it, have a lot of support.
They get into all the post and twist at will, why it is of general interest.
This also, I like to read,
good job.
ZirconiumX wrote:Each game on my octal took about 5 seconds (40/1+0.05), and I can run 7 at once.
100,000 * 5 = 500,000 seconds / 7 = 71,428.6 seconds (call it 80,000 or so)
80,000 / 60 = 1333 minutes
1,333 / 60 = 22 hours
22 hours / 5.5 hours I can run it for = 4 continuous days.
I suggest we release 2.3 now, so we don't wait forever to get everything tuned up.
Is there anywhere else you can think of where we can add a small amount of Elo? One guy on TCEC mentioned a material imbalance table being an Elo loss - we'll have to see what we can do.
One area we can definitely improve on is the code readability, which is pretty poor.
Matthew:out
Matthew,
Some testing is needed even if we have no plan to increase Elo in 2.3, but merely add Linux. Please review the general concept of regression testing, http://en.wikipedia.org/wiki/Regression_testing Cutting down on testing means running a risk of damaging the product's reputation. We have to make sure that 2.3 is at least as good as 2.2.2.
