Off-by-1-error in backtracking?

[Sciss]

Looking at

https://github.com/daoliker/viterbi/blob/master/viterbi_decoder.cc#L35

shouldn't that be

hid_seq[t] = psi[t + 1][hid_seq[t + 1]];

it seems off-by-one for my glance; especially note that this way hid_seq[0] will always be zero which I think is not intended.

[Mary62442]

Hi @Sciss, I have noticed the same thing. The Backtracking equation states that you should search in the psi array at position t+1 and not position t. I have written the Viterbi algorithm in JavaScript and was trying to make a litmus test by checking what other developers have written. A valid implementation in python is here: https://people.eng.unimelb.edu.au/tcohn/comp90042/HMM.py I was about to open an issue as well, but you beat me to it ;)