Changelog¶
v0.2.1¶
An error related to the processing of categorical data with too many categories has been corrected.
Removed
-march=native
flag in the build script to avoid the use of instruction sets not available on some CPUs.
v0.2.0¶
Added conditional linear Gaussian networks (
CLGNetworkType
,CLGNetwork
,ConditionalCLGNetwork
andDynamicCLGNetwork
).Implemented
ChiSquare
(andDynamicChiSquare
) indepencence test.Implemented
MutualInformation
(andDynamicMutualInformation
) indepencence test. This independence test is valid for hybrid data.Implemented
BDe
(Bayesian Dirichlet equivalent) score (andDynamicBDe
).Added
UnknownFactorType
as defaultFactorType
for Bayesian networks when the node type could not be deduced.Added
Assignment
class to represent the assignment of values to variables.
API changes:
Added method
Score.data()
.Added
BayesianNetworkType.data_default_node_type()
for non-homogeneousBayesianNetworkType
.Added constructor for
HeterogeneousBN
to specify a defaultFactorType
for each data type. Also, it addsHeterogeneousBNType.default_node_types()
andHeterogeneousBNType.single_default()
.Added
BayesianNetworkBase.has_unknown_node_types()
andBayesianNetworkBase.set_unknown_node_types()
.Changed signature of
BayesianNetworkType.compatible_node_type()
to include the new node type as argument.Removed
FactorType.opposite_semiparametric()
. This functionality has been replaced byBayesianNetworkType.alternative_node_type()
.Included model as argument of
Operator.opposite()
.Added method
OperatorSet.set_type_blacklist()
. Added a type blacklist argument toChangeNodeTypeSet
constructor.
v0.1.0¶
First release! =).