-
assigned issue to
Pierre Denis
Machine learning for BN with hidden variables
Issue #49
closed
For building Bayesian network by machine learning, Lea provides the build_bn_from_joint method. This method works fine but can only build up BN made up of observed random variables, by means of a joint probability table. In other words, this method cannot produce a BN with "hidden" variables, that are not observed variables (or even not observable).
It would be interesting that Lea provides a method to learn BN with "hidden" variables based on maximum likelihood estimation. The expectation–maximization (EM) algorithm could be used for this purpose. Alongside, Lea should provide a log-likelihood method to provide a measure of the fitting of a model with observed data.
Comments (20)
-
repo owner
-
repo owner
Add learn_by_em, cross_entropy and kl_divergence methods in Lea class (refs
#49)→ <<cset ba6e686e3949>>
-
repo owner
Develop EM algorithm through learn_by_em, em_step and _em_step methods; add Plea, Olea classes for binomial and Poisson distributions (refs
#49)→ <<cset 0c73793ccb8d>>
-
repo owner
Correct doc of Lea class (refs
#49)→ <<cset f178e8a99ddf>>
-
repo owner
Misc corrections in Lea class (refs
#49)→ <<cset f5e7c88a6a33>>
-
repo owner
Move binom and poisson methods in Lea class; add gen_em_steps generator method (refs
#49)→ <<cset 3ddbad1f8507>>
-
repo owner
Add test_em_algo.py for testing EM algorithm (refs
#49)→ <<cset 3d5fa3609b8e>>
-
repo owner
Fix _em_step methods in Flea and Flea2 classes (refs
#49)→ <<cset e4921b94af0d>>
-
repo owner
Simplify Lea.learn_by_em and Lea.gen_by_em signatures; add tolerance arguments to Lea.equiv_f method (refs
#49)→ <<cset 161337363d6e>>
-
repo owner
Correct typo in doc (refs
#49)→ <<cset 8126c410639c>>
-
repo owner
- changed status to closed
-
repo owner
Add learn_by_em, cross_entropy and kl_divergence methods in Lea class (refs
#49)→ <<cset 1571379a262b>>
-
repo owner
Develop EM algorithm through learn_by_em, em_step and _em_step methods; add Plea, Olea classes for binomial and Poisson distributions (refs
#49)→ <<cset 2365f91e4f9a>>
-
repo owner
Correct doc of Lea class (refs
#49)→ <<cset 396fd6773256>>
-
repo owner
Misc corrections in Lea class (refs
#49)→ <<cset 7aa1dbd6ee57>>
-
repo owner
Move binom and poisson methods in Lea class; add gen_em_steps generator method (refs
#49)→ <<cset 6b8078cdf1f9>>
-
repo owner
Add test_em_algo.py for testing EM algorithm (refs
#49)→ <<cset 36ff539da59f>>
-
repo owner
Fix _em_step methods in Flea and Flea2 classes (refs
#49)→ <<cset 4cd49f6ae248>>
-
repo owner
Simplify Lea.learn_by_em and Lea.gen_by_em signatures; add tolerance arguments to Lea.equiv_f method (refs
#49)→ <<cset 4af4777358b0>>
-
repo owner
Correct typo in doc (refs
#49)→ <<cset ea478461cb29>>
- Log in to comment
