markovchain
markovchain copied to clipboard
Hitting probabilities function should allow user to select target states and solver method
The hitting probabilities function is very slow for large markov chains. Users should have the option to select the target states so that unneeded values aren’t computed.
Additionally, directly solving this system of equations is impractical in many cases. Users should have the option to use numerical methods to solve for hitting (and absorption) probabilities. I have found that biconjugate gradient descent works very well for this purpose and can be many orders of magnitude faster than directly solving, even when the identity matrix is used as a preconditioner.
Thank for the note. Unfortunately I'm not be able to work on this topic are you willing to propose a pull request?
Il mar 15 giu 2021, 21:54 EliSmith45 @.***> ha scritto:
The hitting probabilities function is very slow for large markov chains. Users should have the option to select the target states so that unneeded values aren’t computed.
Additionally, directly solving this system of equations is impractical in many cases. Users should have the option to use numerical methods to solve for hitting (and absorption) probabilities. I have found that biconjugate gradient descent works very well for this purpose and can be many orders of magnitude faster than directly solving, even when the identity matrix is used as a preconditioner.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/spedygiorgio/markovchain/issues/203, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAGJNRSOFXFWMWZXEPT2TDTTS6VVXANCNFSM46X57G6A .
@spedygiorgio No problem, I’ll propose a pull request after I pretty up my code a bit.