Forward–backward algorithm: Difference between revisions

Revision as of 12:04, 12 October 2020

Revision as of 13:25, 21 October 2020

Line 349:

prev_f_sum = start_prob[st]

else:

prev_f_sum = sum(f_prev[k] * trans_prob[k][st] for k in states)

f_curr[st] = emm_prob[st][observation_i] * prev_f_sum

Line 360:

# Backward part of the algorithm

bkw = []

for i, observation_i_plus in enumerate(reversed(observations[1:] + (None,))):

b_curr = {}

for st in states:

Line 381:

assert p_fwd == p_bkw

return fwd, bkw, posterior

</syntaxhighlight>