I am trying to understand the Baum Welch algorithm by implementing it in xls. I have chosen a simple example of observations from a loaded (L) vs fair (F) die. I calculate the forward and backward probabilities for L and F states. I then calculate the probability of observations. I know the calculations till this point are correct, since i get the same p(observations) for each row of data.

i.e.p(observations)= forward probability(L)*forward probability(F)+ backward probability(L) * backward probability(F) is same for all rows as it should be

Then I proceed to calculate the transition probabilities. Here I find that the value of the transition probability for the state transition L->L is coming to be above 1.For other states the values seem to be correct

I calculate the transition probabilities for L->L for each row as follows: T(L->L)

(Forward_probability(L,t)* Transition(L->L)* emission(observation|L) * backward_probability(L,t+1))/ probability(observations)

The estimated probability = sum( T(L->L) )/sum(L|observation ) I am unable to find why my transition probability has value > 1.

Any help will be appreciated.


1 Answer 1


I found out the issue. I was using the emission probability of current observation whereas I should have used for t+1 th observation. I get correct answer now

  • $\begingroup$ Congratulations on resolving your issue. You're not the first one to have broken the unity barrier in probability, and with probability $>$ 1, you won't be the last. But unlike some, at least you realized you had probabilities exceeding one. $\endgroup$ May 12, 2016 at 0:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.