I am using a code that calls the hmmviterbi hundreds of thousands of times for long sequences of emissions (about 10000000 symbols) and big A and B probabilities matrices at each call. This makes my code runs for days.
Is there any other implementation, than that offered by the HMM toolbox, that runs faster?
Here the function called hundreds of thousands of times:
function [ acc ] = fitness(A, B) hmm_states = hmmviterbi(testing_data.getsetData.seqs,A,B) ; c = hmm_states == testing_data.getsetData.states ; acc = ( sum(c(:) == 1) ) / size(hmm_states,2) ; end
Thank you in advance.
Dear Amer,
Use "Parallel pool" in Matlab for doing analysis simultaneously.
Regards
- S. M.
One simple way is to convert it to a C++ MEX file and call it inside your MATLAB code.
Dear Amer Draa,
Look the link, maybe useful.
https://www.mathworks.com/matlabcentral/fileexchange/24516-forward-viterbi-algorithm
https://stats.stackexchange.com/questions/274961/matlab-hmm-viterbi-algorithm-calculated-state-sequence-probability
Regards,
Shafagat
Dear Amer Draa ,
Instead of using a big data, you can use a data reduction method such as DWT, DCT and PCA.
Dear Amer,
Easily to convert file.m to file.mex, just follow the steps in the following link:
- https://www.mathworks.com/matlabcentral/mlc-downloads/downloads/submissions/47593/versions/20/previews/documentation/modules/CodeGeneration/html/AccelerateMATLABDocumentation.html
- Badges
- Science topic