You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hey there,
I am trying to unserstand how to use Dirichlet priors on the parameters of a
CPT. In the documentation I can see how to define a dirichlet prior, but it
seems to me that it is not possible to specify the Dirichlet pseudo counts
separately for each entry.
In my case, I have a prior CPT that represents my prior knowledge, but if I
define a dirichlet prior over my parameters, I can only specify the ESS.
The function bayes_update_params will then recompute the pseudo counts by
summing the counts present in the data for a particular parent-child
combination with the dirichlet prior counts, but neglecting the prior CPT.
If I have that P(X=x | Pa1 = p) with probability 0.9 and a BDeu Dirichlet prior
with ESS = 10
I would expect my dirichlet prior to already have a pseudo count of 9 for that
particulare state, while instead I get the same number of pseudo counts spread
over all the possible states.
The results of this is that my posterior is being relatively flat and drive by
the data, despite my prior CPT being sharply peaked around some entries.
What am I doing wrong?
Original issue reported on code.google.com by [email protected] on 25 Oct 2011 at 1:02
The text was updated successfully, but these errors were encountered:
Original issue reported on code.google.com by
[email protected]
on 25 Oct 2011 at 1:02The text was updated successfully, but these errors were encountered: