If I can answer, Persistent Hash is just a method of saving the current Hash Table to the hard drive or SSD. People wishing to analyze multiple positions can save this information from one position, analyze another position, then reload the first position and Hash to continue analysis later.
Correspondence have asked for this feature. Please note the files can be huge if your Hash table size is large, and that a given hash table will not have the whole search tree in it, but if you have spent hours analyzing a position, it will save time some other day if you want the program to examine it more.
mjlef wrote:If I can answer, Persistent Hash is just a method of saving the current Hash Table to the hard drive or SSD. People wishing to analyze multiple positions can save this information from one position, analyze another position, then reload the first position and Hash to continue analysis later.
Correspondence have asked for this feature. Please note the files can be huge if your Hash table size is large, and that a given hash table will not have the whole search tree in it, but if you have spent hours analyzing a position, it will save time some other day if you want the program to examine it more.
Hi Mark,
A learning file similar to what Houdini or Jeremy B.'s Stockfish had may be an even better feature to have. Such a file usually stays small and contains only useful evals from higher depths, based on a threshold set by the user.
Permanent hash files tend to be big and bulky and not worth keeping in the long run, unlike a learning file.
Yes, I had a type of learning file in my old program NOW, but I have not had time to work on it for Komodo yet. But it is on the list. There are a lot of things to consider about what should be kept and what should be tossed out in a "learning" system. Since the program has to reload the "learned" positions before every search, you want to keep the files small, so getting it right is important. Persistent Hash is indeed much simpler, so like any lazy person, I tend to grab the easy fruit first!
We appreciate the feedback, and the more we hear that a lot of people want a feature, the more likely we will give it a higher priority.
mjlef wrote:If I can answer, Persistent Hash is just a method of saving the current Hash Table to the hard drive or SSD. People wishing to analyze multiple positions can save this information from one position, analyze another position, then reload the first position and Hash to continue analysis later.
Correspondence have asked for this feature. Please note the files can be huge if your Hash table size is large, and that a given hash table will not have the whole search tree in it, but if you have spent hours analyzing a position, it will save time some other day if you want the program to examine it more.
I see...
I'm glad I asked because I was thinking of something a little different. But, I agree that this would be very helpful for CC or any long analysis.
I would also be interested in the learning type as well.
Regards,
Zen
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.
I'd like to request the learning file type of learning as well, which I think is what Hiarcs and Shredder have used for a long time. With this you can for instance play out a game from a position you are interested in, then go back to the position and reanalyze with the saved learning being used.
Werewolf wrote:Bearing in mind that LP is just a speedup and nothing more, you have to factor in the time it takes to set it up in the first place....
Run Arena as administrator, large pages then will be available for all engines that support them.
It's also very easy to verify LP is working. The engines will report they are using LP, and the task manager can also be used to verity.
I don't know what other "set up" is needed (??). And yes, it is "just a speedup", but IMO 15% is worth it. People have spent way more effort to get compiles only a few % speedup.
Anyway, maybe team Komodo will consider implementing a large page feature one day.....pretty please! But 2 out of 3 ain't bad for my wishlist with K9 having FRC and persistent hash.
PS: At present I don't want to fiddle with linux install, as was also suggested, but thanks for that info Syzygy de Man.
Good pricing model, makes the subscription model compelling - get at least one, maybe two extra versions a year. Question on the subscription model - what happens after the year is up - and you do not resubscribe. Will the engines you have stop working?
MikeB wrote:Good pricing model, makes the subscription model compelling - get at least one, maybe two extra versions a year. Question on the subscription model - what happens after the year is up - and you do not resubscribe. Will the engines you have stop working?
No, you just don't get future upgrades without buying them.