Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task manager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task manager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
In my practical test, 49 mins of infinite analysis gave me a situation where I still had heavy fails high and low, so I couldn't be sure of the result. 5-10 mins of hash load + 40 mins of anaòysis gave me about +400 on the ending, being sure to win.
Did you try to set hashes to 2 GBs as I suggested before?
Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task mlanager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
I am having problems too in reproducing Rodolfo efforts under scid on Ubuntu. I tried both native compilation and execution through wine, but apparently the .hsh file is not generated.
Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task mlanager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
I am having problems too in reproducing Rodolfo efforts under scid on Ubuntu. I tried both native compilation and execution through wine, but apparently the .hsh file is not generated.
I don't know under Linux, but I can say you should run the GUI as administrator under Windows. If not, engine can't write anything.
retep1 wrote:
which search depth: 20 or 30 or what?
With which tool do you see the disk accesses under Windows?
About 5 seconds, doesn't matter the depth. On Arena I see the hash size growing and when it reaches 2 GBs I stop infinite analysis, I go to UCI options, I select the file to load and I click "LoadHashToFile".
The "tool" I use is... the green led of disk access.
Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task mlanager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
I am having problems too in reproducing Rodolfo efforts under scid on Ubuntu. I tried both native compilation and execution through wine, but apparently the .hsh file is not generated.
Finally, I managed to reproduce Rodolfo Stockfish_savehash findings in Arena on my linux machine. I found that there is no need to be root/administrator to save the hash table on disk. Also, the two GB table loaded quickly without the need of prerunning the search.
My first impression is that this can become a tool that can approach the utility of Aquarium IDEA, a program widely used by top correpondence top-player or by anyone looking for the truth in any position. Of course, as shown by Rodolfo, users must be willing to induce learning by interactively playing with the tool. To this purpose I was wondering if it is possible to export or convert the Transposition Table to a book bin format for offline viewing. Probably, also exporting the TT to epd positions will do the trick.
I hope to post myself in the future some examples of usage of Stockfish_savehash. In the meantime, I would really thank Rodolfo and Daniel for their contribution to the fascinating field of computer learning.
Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task mlanager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
I am having problems too in reproducing Rodolfo efforts under scid on Ubuntu. I tried both native compilation and execution through wine, but apparently the .hsh file is not generated.
Finally, I managed to reproduce Rodolfo Stockfish_savehash findings in Arena on my linux machine. I found that there is no need to be root/administrator to save the hash table on disk. Also, the two GB table loaded quickly without the need of prerunning the search.
My first impression is that this can become a tool that can approach the utility of Aquarium IDEA, a program widely used by top correpondence top-player or by anyone looking for the truth in any position. Of course, as shown by Rodolfo, users must be willing to induce learning by interactively playing with the tool. To this purpose I was wondering if it is possible to export or convert the Transposition Table to a book bin format for offline viewing. Probably, also exporting the TT to epd positions will do the trick.
I hope to post myself in the future some examples of usage of Stockfish_savehash. In the meantime, I would really thank Rodolfo and Daniel for their contribution to the fascinating field of computer learning.
In my PC, engine doesn't load hashes unless I start few seconds of infinite analysis. It could depend by the GUI or OS, I don't know. A 2 GBs hash takes 5 to 10 minutes to load, here. It could be faster if I keep the hash on a flash memory card. I didn't try but it could be a good solution.
I remember Aquarium IDeA as an automated search. It has the limit it can ignore a brilliant move if its first search wasn't one of the best, then I guess it's not the best possible tool. A human has the advantage he can try not probable moves too, if his istinct says him there could be something interesting there. Human skill plays a role with this kind of analysis because it's not only the fact of following the PV and taking moves back.
Currently, there's a project about a dedicated GUI for analysis with PHs. Engine requisite to operate with that GUI is to accept an "import epd" command. I don't want to anticipate too much about it. If interest about persistent hashes and their use will grow that'll be released. If not, that'll remain a private project.
Rodolfo Leoni wrote:I'm getting a bit tired of being the only one running tests on this hash system. I'll wait to see someone's else effort to contribute before continuing.
Unfortunately, I'm not able to do PH "loadHashFromFile" in one of the GUIs (Fritz 15, Chessbase 14, Aquarium 2017 or Arena 3.5.1) Save works, but do not load. In the task mlanager (resource monitor), hard disk accesses are visible when saving, but not when loading.
Therefore, I can contribute nothing to the test.
But still a remark to the benefit. If the loading of 2 GB hash takes about 5 to10 minutes, where is the time advantage compared to the infinite analysis from scratch. How big was the search-depth?
I am having problems too in reproducing Rodolfo efforts under scid on Ubuntu. I tried both native compilation and execution through wine, but apparently the .hsh file is not generated.
Finally, I managed to reproduce Rodolfo Stockfish_savehash findings in Arena on my linux machine. I found that there is no need to be root/administrator to save the hash table on disk. Also, the two GB table loaded quickly without the need of prerunning the search.
My first impression is that this can become a tool that can approach the utility of Aquarium IDEA, a program widely used by top correpondence top-player or by anyone looking for the truth in any position. Of course, as shown by Rodolfo, users must be willing to induce learning by interactively playing with the tool. To this purpose I was wondering if it is possible to export or convert the Transposition Table to a book bin format for offline viewing. Probably, also exporting the TT to epd positions will do the trick.
I hope to post myself in the future some examples of usage of Stockfish_savehash. In the meantime, I would really thank Rodolfo and Daniel for their contribution to the fascinating field of computer learning.
In my PC, engine doesn't load hashes unless I start few seconds of infinite analysis. It could depend by the GUI or OS, I don't know. A 2 GBs hash takes 5 to 10 minutes to load, here. It could be faster if I keep the hash on a flash memory card. I didn't try but it could be a good solution.
I remember Aquarium IDeA as an automated search. It has the limit it can ignore a brilliant move if its first search wasn't one of the best, then I guess it's not the best possible tool. A human has the advantage he can try not probable moves too, if his istinct says him there could be something interesting there. Human skill plays a role with this kind of analysis because it's not only the fact of following the PV and taking moves back.
Currently, there's a project about a dedicated GUI for analysis with PHs. Engine requisite to operate with that GUI is to accept an "import epd" command. I don't want to anticipate too much about it. If interest about persistent hashes and their use will grow that'll be released. If not, that'll remain a private project.
Let us see!
Thanks for the update, Rodolfo. I would be greatly itnterested in knowing about your project. Keep up the good work!
Currently, there's a project about a dedicated GUI for analysis with PHs. Engine requisite to operate with that GUI is to accept an "import epd" command. I don't want to anticipate too much about it. If interest about persistent hashes and their use will grow that'll be released. If not, that'll remain a private project.
Such a GUI would be a great step towards a better analysis of positions in a chess game - much success in developing this non-private GUI and, of course, I have a great interest in this GUI
