bnlearn

• Latest version of pandas is supported again.

• Fix for showing the weights in the edges for the static plot

• Fix for issue #99
• Removed d3blocks from setup /requirements and added it as check when using the interactive plot functionality

• add d3blocks to setup and requirements with minimum version. The wrong version can break the interactive plots.

• Fix the issue, sorting the continuous_columns in the correct order when discretizing continuous data @ankh1999

• Fix for the dynamic plot. #89 The labels are now correctly shown again.
• Fix for when calling the .predict method. #87 Many thanks to @ankh1999

• fix for recursive model check

• Removed lowering of pathname in the import_DAG function.

• Added new scoring method: BDS and AIC

• Created groupby functionality for inferences that allows grouping similar categories.

• Auto-convert non-numerical columns into numerical when discretizing

• Code refactoring

• Small improvements docstrings

I am really happy to announce the following new great functionalities in bnlearn!

• Continuous data modeling is possible using the approach of Yi-Chun Chen et al.
• Interactive plots are created using D3Blocks which provides more flexibility.
• Data examples and functions are removed from bnlearn and are now imported from datazets library.
• Unit tests updated and new ones implemented

Especially for the plot functionality, some of the input parameters are changed.
See the docs for continuous data modeling

• fix for disabling verbose message

• Freeze pandas package to 1.5.3 because of error in iteritems that is used in underlying pgmpy package.
• A huge new implementation with new functionalities but this needs to be documented first.

• added library requests

• Fix for import examples (requires latest numpy version)
• replaced wget with internal python function.

• Added cdn resources to fix interactive plot in jupyter notebook and Colab

• Fix for interactive plot. It did not show up correctly but now it does again.

• sklearn is changed to scikit-learn in the setup and requirements (issue #66 )

• plot function bn.plot(model) return figure axis.
• Setup file contains import of python-louvain instead of community
• docstring updates regarding this output.

Example:

model = bn.import_DAG('sprinkler', CPD=True)
df = bn.sampling(model, n=1000, methodtype='bayes')
fig = bn.plot(model)
fig['ax']

• Added functionality to compute sctructure scores when using parameter of structure learning. This can help to decide which scoring type may give a better fit ('bic', 'k2', 'bdeu'). The results scores are directly stored in the output.

import bnlearn as bn

# Load example dataset
df = bn.import_example('sprinkler')

edges = [('Cloudy', 'Sprinkler'),
         ('Cloudy', 'Rain'),
         ('Sprinkler', 'Wet_Grass'),
         ('Rain', 'Wet_Grass')]

# Make the actual Bayesian DAG
DAG = bn.make_DAG(edges)
model = bn.parameter_learning.fit(DAG, df)

model['structure_scores']
{'k2': -33962.61414408797,
 'bds': -57992.919156623604,
 'bic': -94337.69274492635,
 'bdeu': -33670.95375881856}

• check_model now uses Decimals to prevent Floating Point Errors in the checks: Issue #60.

• CPDs are now returned from print_CPD for further usage.