bnlearn

• Fix for issues #37 where query2df is fixed to output a structured dataframe with the query information.
• Added unit test for this issue.

• Added Gibbs sampling. See the docs for examples.
• Improvements in verbosity messages.
• Some code refactoring/cleaning.

• added input parameter n_jobs where possible

• improved speed in predict functionality
• Add scoretype to parameter_learning

• smooth parameter added to parameter_learning

import bnlearn as bn
DAG = bn.import_DAG('water', verbose=0)
# Sampling
df = bn.sampling(DAG, n=1000)
# Parameter learning
model = bn.parameter_learning.fit(DAG, df, smooth=None)

• Fix for plotting due to new version of networkx. The version of networkx should be >= 2.7.1
• Layout can now be changed more easily
• figsize added as an input parameter which is more convenient than width/height.

import bnlearn as bn
df = bn.import_example('asia')
model = bn.structure_learning.fit(df)
# Plot
bn.plot(model)
# Plot with different layouts
bn.plot(model, params_static={'layout':'spectral_layout'})
bn.plot(model, params_static={'layout':'planar_layout'})
bn.plot(model, params_static={'layout':'kamada_kawai_layout'})
bn.plot(model, params_static={'layout':'spring_layout'})
bn.plot(model, params_static={'layout':'circular_layout', "figsize": (15, 10)})

• added pyvis to setup file and requirements.

• fix for bn.print_CPD()
• name convention: bayesianmodel into bayesiannetwork

• Name convention: BayesianModel is changed into BayesianNetwork

• New conversion functionalities to convert source-target vector into sparse dataframe: vec2df
• Sphinx pages updated: https://erdogant.github.io/bnlearn/pages/html/dataframe%20conversions.html
• Docstrings updated and included.

import bnlearn as bn
# Load large example with source-target edges
vec = bn.import_example("stormofswords")
# Convert to dataframe
df = bn.vec2df(vec['source'], vec['target'], weights=vec['weight'])

• Added new functionality vec2df to create dataframe from source-target-weights
• Added new example to demonstrate the usage
• updated some docstrings

Example vec2df:

import bnlearn as bn

source=['Cloudy','Cloudy','Sprinkler','Rain']
target=['Sprinkler','Rain','Wet_Grass','Wet_Grass']
weights=[1,2,1,3]

# Convert into sparse datamatrix
df = bn.vec2df(source, target, weights=weights)

#    Cloudy   Rain  Sprinkler  Wet_Grass
# 0    True  False       True      False
# 1    True   True      False      False
# 2    True   True      False      False
# 3   False  False       True       True
# 4   False   True      False       True
# 5   False   True      False       True
# 6   False   True      False       True

• Improvements in independence test. It should also work now after parameter_learning.
• Improvements in plots
• Code refactoring

• implementation of bnlearn.independence_test functionality that allows to compute edge strength
• plots improved for the edge weights
• more pep styling
• some code refactoring

# Example:

# Load asia DAG
df = bn.import_example(data='alarm')
# Structure learning of sampled dataset
model = bn.structure_learning.fit(df)
# Compute edge strength with the chi_square test statistic. Set prune=True to remove the none-significant edges.
model = bn.independence_test(model, df, test='chi_square', prune=False)
# Make plot
bn.plot(model)

• small fixes

• Bugfix when plotting when using edge_properties.
• added naivebayes method for creating DAG.
• added naivebayes method for structure learning.
• code refactoring.
• Updating docstrings.
• Updated unit tests

Let me know if there are accidental new bugs introduced!

• edge properties can be changed for plotting purposes
  • color (per edge)
  • weight (per edge)
  • edge_alpha (overall)
  • arrowstyle (overall)
  • arrowsize (overall)

• added doi

• added missing package: tabulate

• output of inference variables is ordered in the same manner as defined in the input.

• Bug fix that fixes handling multiple categorical values in the inference.