sisl / BayesNets.jl / 184 / 2
85%
master: 85%

DEFAULT BRANCH: master
Ran
Files 37
Run time 4s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 87.698% (+12.9%) from 74.796%
184.2

push

travis-ci

web-flow 
Inference (#52)

* Outlined inference methods. Added .gitignore

Added gitignore to ignore vim swp files.
New file inference.jl for exact, likelihood, gibbs, and loop belief (hopefully)
Added include to BayesNets (and newline).

* Added random bayesnet, fixed likelihood.

Changed likelihood to be more efficient than before.
Added functions to generate bayesnets.

* Started Gibbs Sampling

Added gibbs sampling method. Sort-of works, but huge memory complexity.

* New sumout, different versions, waiting on lbp

Have yet to add LBP.
Tww versions of likelihood: pre-allocated and not
Two versions of gibbs: full iteration and not
Changed factors.jl summout to work with non-binary variables.
    (Hopefully its a valid change)

* Moved unnecessary code to new file

Code to generate BaeyesNets was moved to gen_bayes_nets.jl

* Zero parent nodes, random inf, inference changes

Now random bayes nets can have nodes with no parents
Added method to generate random inference scenario

* More streamlined sumout and normalize

Normalize uses builtin broadcasting
Summout takes in both vectors and symbols
Fixing problems with exact inference

* It works!! Maybe??

Still need to fully test. LBP may be implemented

* Moved methods to separate files

All inference methods are in their own files
Added an ncategories(bn, symbol) method to simplify code.
Keeping original inference.jl file for now . . .

* Final commit.

Finalized all the algorithms, prettied up the code some, deleted the original
file and added some more comments.

* Porting over inference to Factors.jl

* Inference interface, sketched out likelihood

Added unified inference interface. Tentative move of exact and likelihood
over.

* Renamed old factors, added new factors.

Deleted other gibbs method.
Added factors as "submodule" instead of separate package.
Changing inference API.
Still no docs of Factors

* Factors pass tests. Still waiting on join.

Integrated Factors into BayesNets. Mostly
Undeleted gibbs since its important and useful
More work on Inference interface to follow
Can have only root node be parentless in rand_disc_bn

* Still working on inference. Exact is broken.

Exact inference using Factors occasionally is completely wrong.
Still working on a debugging

* Join fully works; exact inf is done (hopefully).

Fixed logical error with join. Added tests (still need work).
Added alias for union NodeNode or Vector of NodeNames
Small change to tables sumout
Renamed probability to potential, because ... I don't know

* Rough sketch of LBP

Started to convert LBP. Still using dicts. Now, not in parallel, but updating
each node one at a time, with previous changes in that iteration factored in.
This likely will not work ...

* LBP outline finished. Still issues with inf

Gibbs and LBP still not providing correct (or senesical) results.
Changes to (Gibbs)Inference object and associated tests.

* LBP works. Very slow though

More work to follow (maybe) on speeding up broadcast/join/reduce for factors.

* Most everything works.

Tests still missing edge cases. Broadcast, reduce, and join are still
incredibly slow. ¯\_(ツ)_/¯

* Test pass. Docs updated. Changed innerworking.

Changes how argument checking is done.
Added convenience method for ncategories of cpd
Updated ipynb docs
Renamed (again) inference/ to Inference.

* stashing

* inference methods incorporated

1383 of 1577 relevant lines covered (87.7%)

5552.08 hits per line

Source Files on job 184.2