Learn file
Moderators: Watchman, Mark Uniacke, mrudolf
Learn file
Hello!
My question is about the lrn.- file, hiarcs creates by himself. It seems to me the engine stores certain chess- positions there, for example of tricky problems to solve, which by this way can be solved much faster the second time even if you transfer the file from one PC to another, which did not solve the prob before. Has this anything to do with bitboards, how much of these positions will be stored and how often by which criteria, is it overwritten sometimes and how big can it get.?
Can I take any influence on it except by selecting or deselection the option position learning?
Best regards
Peter.
My question is about the lrn.- file, hiarcs creates by himself. It seems to me the engine stores certain chess- positions there, for example of tricky problems to solve, which by this way can be solved much faster the second time even if you transfer the file from one PC to another, which did not solve the prob before. Has this anything to do with bitboards, how much of these positions will be stored and how often by which criteria, is it overwritten sometimes and how big can it get.?
Can I take any influence on it except by selecting or deselection the option position learning?
Best regards
Peter.
- Mark Uniacke
- Hiarcs Author
- Posts: 1458
- Joined: Sun Jul 29, 2007 1:32 pm
- Location: United Kingdom
- Contact:
Re: Learn file
The learn file is not related to bitboards.peter wrote:Hello!
My question is about the lrn.- file, hiarcs creates by himself. It seems to me the engine stores certain chess- positions there, for example of tricky problems to solve, which by this way can be solved much faster the second time even if you transfer the file from one PC to another, which did not solve the prob before. Has this anything to do with bitboards, how much of these positions will be stored and how often by which criteria, is it overwritten sometimes and how big can it get.?
Can I take any influence on it except by selecting or deselection the option position learning?
Best regards
Peter.
Hiarcs uses a unique criteria to decide which positions to store in the learn file. I cannot easily explain the criteria because of the internal detail of Hiarcs is rather unique, but in essence Hiarcs is looking for the unexpected and if it finds it then that position becomes more interesting for the future.
The position learning option decides whether Hiarcs will try to learn from the games it is playing.
-
- Member
- Posts: 21
- Joined: Thu Aug 02, 2007 1:48 am
- Location: Idyllwild, California
About using .lrn files from previous versions?
I ran HIARCS 11.1 for quite a while, and HIARCS 10 for a much longer time before that. When I installed HIARCS 11.2, there was no .lrn file yet. Would it help anything to put one or more of these earlier .lrn files in the same folder with HIARCS 11.2? With or without renaming it to fit 11.2? Thanks. -Jack
Jack Clark
Idyllwild, California
Idyllwild, California
- Harvey Williamson
- Site Admin
- Posts: 6079
- Joined: Sun Jul 29, 2007 6:57 am
- Location: Media City, UK
- Contact:
Re: About using .lrn files from previous versions?
Hi Jack,Jack Clark wrote:I ran HIARCS 11.1 for quite a while, and HIARCS 10 for a much longer time before that. When I installed HIARCS 11.2, there was no .lrn file yet. Would it help anything to put one or more of these earlier .lrn files in the same folder with HIARCS 11.2? With or without renaming it to fit 11.2? Thanks. -Jack
11.2 will create a .lrn file the 1st time you use it. You can probably rename your 11.1 .lrn as 11.2 and it should work. I dont think the older ones will.
Best Wishes,
Harvey
Re: Learn file
I may have made this suggestion a while back, but it would be nice to have an option so that learning occurs more often, this would be especiallyMark Uniacke wrote:The learn file is not related to bitboards.peter wrote:Hello!
My question is about the lrn.- file, hiarcs creates by himself. It seems to me the engine stores certain chess- positions there, for example of tricky problems to solve, which by this way can be solved much faster the second time even if you transfer the file from one PC to another, which did not solve the prob before. Has this anything to do with bitboards, how much of these positions will be stored and how often by which criteria, is it overwritten sometimes and how big can it get.?
Can I take any influence on it except by selecting or deselection the option position learning?
Best regards
Peter.
Hiarcs uses a unique criteria to decide which positions to store in the learn file. I cannot easily explain the criteria because of the internal detail of Hiarcs is rather unique, but in essence Hiarcs is looking for the unexpected and if it finds it then that position becomes more interesting for the future.
The position learning option decides whether Hiarcs will try to learn from the games it is playing.
useful for autoplaying from interesting openings and startup positions!
Andrew
-
- Member
- Posts: 2
- Joined: Thu Aug 30, 2007 4:18 pm
Re: About using .lrn files from previous versions?
Hello!Harvey Williamson wrote:Hi Jack,Jack Clark wrote:I ran HIARCS 11.1 for quite a while, and HIARCS 10 for a much longer time before that. When I installed HIARCS 11.2, there was no .lrn file yet. Would it help anything to put one or more of these earlier .lrn files in the same folder with HIARCS 11.2? With or without renaming it to fit 11.2? Thanks. -Jack
11.2 will create a .lrn file the 1st time you use it. You can probably rename your 11.1 .lrn as 11.2 and it should work. I dont think the older ones will.
Best Wishes,
Harvey
I plan to upgrade my currently used HIARCS to the newest HIARCS 11.2 version at this web site within the next days.
Now my question about the learning file with HIARCS 11.2:
Since there is a lot of computing power sticking in the old learn file from my previous version I would like to know if the information above, that older versions of learning files will not work with the current version of HIARCS, is proven to be correct and if so, if there is some converter to transform an older learning file to the newset version or if there is going to upcome some converter in near future for this purpose?
Thank you for any information about this
- Harvey Williamson
- Site Admin
- Posts: 6079
- Joined: Sun Jul 29, 2007 6:57 am
- Location: Media City, UK
- Contact:
Re: About using .lrn files from previous versions?
Hi,Sebastian Andreas Gruber wrote:Hello!Harvey Williamson wrote:Hi Jack,Jack Clark wrote:I ran HIARCS 11.1 for quite a while, and HIARCS 10 for a much longer time before that. When I installed HIARCS 11.2, there was no .lrn file yet. Would it help anything to put one or more of these earlier .lrn files in the same folder with HIARCS 11.2? With or without renaming it to fit 11.2? Thanks. -Jack
11.2 will create a .lrn file the 1st time you use it. You can probably rename your 11.1 .lrn as 11.2 and it should work. I dont think the older ones will.
Best Wishes,
Harvey
I plan to upgrade my currently used HIARCS to the newest HIARCS 11.2 version at this web site within the next days.
Now my question about the learning file with HIARCS 11.2:
Since there is a lot of computing power sticking in the old learn file from my previous version I would like to know if the information above, that older versions of learning files will not work with the current version of HIARCS, is proven to be correct and if so, if there is some converter to transform an older learning file to the newset version or if there is going to upcome some converter in near future for this purpose?
Thank you for any information about this
Can you tell me exactly what version you are upgrading from?
Thanks,
Harvey
-
- Member
- Posts: 2
- Joined: Thu Aug 30, 2007 4:18 pm
Hello Harvey Williamson!
Thank you for your fast reply!
Actually I will upgrade from the version 9, but my most important learning file might be the one I'm still using from version 8.
Since I play correspondance chess with some friends for years now, as well as unregular play at IECG and I do analysis on scandinavian (B01) for almost 5 years now, I am not sure if and how important the learning file has become at this time.
I do not fully understand what weight a learning file becomes over the years if it does so. Perhaps it's not as important as I might expect about it could be?
But it would be nice if it could be possible to reuse learning files from older versions over the years and different versions of HIARCS I guess so someone might gain additional advantage from their existance in long term?
Do you think there it is possible to convert older learning files in some way or simply rename them to newer named versions or if there is anything planed to make it possible to have reusable learning files in future of HIARCS?
Apart from this question I would be pleased to know how strong you would expect the influence of learning files when trained over a longer peroid of time to have a better understanding how important they are to engine play. Do you have any general information about this please?
Thank you!
Thank you for your fast reply!
Actually I will upgrade from the version 9, but my most important learning file might be the one I'm still using from version 8.
Since I play correspondance chess with some friends for years now, as well as unregular play at IECG and I do analysis on scandinavian (B01) for almost 5 years now, I am not sure if and how important the learning file has become at this time.
I do not fully understand what weight a learning file becomes over the years if it does so. Perhaps it's not as important as I might expect about it could be?
But it would be nice if it could be possible to reuse learning files from older versions over the years and different versions of HIARCS I guess so someone might gain additional advantage from their existance in long term?
Do you think there it is possible to convert older learning files in some way or simply rename them to newer named versions or if there is anything planed to make it possible to have reusable learning files in future of HIARCS?
Apart from this question I would be pleased to know how strong you would expect the influence of learning files when trained over a longer peroid of time to have a better understanding how important they are to engine play. Do you have any general information about this please?
Thank you!
- Harvey Williamson
- Site Admin
- Posts: 6079
- Joined: Sun Jul 29, 2007 6:57 am
- Location: Media City, UK
- Contact:
- Mark Uniacke
- Hiarcs Author
- Posts: 1458
- Joined: Sun Jul 29, 2007 1:32 pm
- Location: United Kingdom
- Contact:
Yes, while the learn files may contain some useful data from your previous games the latest engine has improved immensely since Hiarcs 8/9 and so will often play into superior lines and find better evaluations quicker.
Also the learn data changed at Hiarcs 11.1 and is not compatible with older versions anymore.
Also the learn data changed at Hiarcs 11.1 and is not compatible with older versions anymore.
Hello!
May I ask for some more details about learnfile, if they can be given?
It would be especially useful for me to know if there is any time limit, Hiarcs needs to estimate positions on the board worth storing at learnfile. So if I for example go back in a game of interest quickly, is there any chance to so to say teach the .lrn?
If so, does there have to be a certain kind of jump of evaluation during the backward analysis?
Third question: is there any limit of data volume and about how big can it get?
Knowing I ask for delicate details many preestimating thanks!
Peter.
May I ask for some more details about learnfile, if they can be given?
It would be especially useful for me to know if there is any time limit, Hiarcs needs to estimate positions on the board worth storing at learnfile. So if I for example go back in a game of interest quickly, is there any chance to so to say teach the .lrn?
If so, does there have to be a certain kind of jump of evaluation during the backward analysis?
Third question: is there any limit of data volume and about how big can it get?
Knowing I ask for delicate details many preestimating thanks!
Peter.
- Mark Uniacke
- Hiarcs Author
- Posts: 1458
- Joined: Sun Jul 29, 2007 1:32 pm
- Location: United Kingdom
- Contact:
The learnt data is prioritised and it tries to keep the most relevent learnt data. This is clearly an on-going process.peter wrote:Hello!
May I ask for some more details about learnfile, if they can be given?
It would be especially useful for me to know if there is any time limit, Hiarcs needs to estimate positions on the board worth storing at learnfile. So if I for example go back in a game of interest quickly, is there any chance to so to say teach the .lrn?
If so, does there have to be a certain kind of jump of evaluation during the backward analysis?
Third question: is there any limit of data volume and about how big can it get?
Knowing I ask for delicate details many preestimating thanks!
Peter.
You can teach the program by progressing along variations which it misunderstands effectively helping to reveal the truth of the position. When Hiarcs recognises it has under or over estimated its evaluation of the possibilities it is able to make improvements so when you back up the line of play or revisit the position it can hopefully make a better decision.
The data set is limited but given its prioritised nature it rarely has any adverse effect, in fact there are some benefits to keep it limited.