Releases: fracpete/python-weka-wrapper3
Releases · fracpete/python-weka-wrapper3
Release v0.2.16
Release v0.2.15
- the methods
create_instances_from_lists
andcreate_instances_from_matrices
of theweka.core.dataset
module can handle missing values now (None
in case of lists,nan
in case of matrices), as well as being able to force columns to be nominal - added the method
load_csv_file
to the moduleweka.core.converters
to provide a more reliable way of loading CSV files compared to Weka's nativeCSVLoader
converter (uses Python's csv module and then callscreate_instances_from_lists
). - added
simple_range
library as dependency to make it easier to generate integer list from range strings - added
weka.core.utils
module with following methods: correlation, variance, normalize - updated requirements (and install instructions) to make pww3 work with Python 3.11+
Release v0.2.14
- extracted arpack_combined.jar, core.jar and mtj.jar from weka.jar to make them available under Java 17
Release v0.2.13
- moved
Configurable
andJSONObject
into configurable-objects library - moved base flow components into simple-data-flow library
- added methods
subset_results
,rank_results
to classAttributeSelection
(module:weka.attribute_selection
) to give access to cross-validation output in numeric rather textual form (NB: it has to parse the textual CV output). - made the
plot_experiments
method (moduleweka.plot.experiments
) more customizable
Release v0.2.12
- method
install_packages
(module:weka.core.packages
) now no longer interprets the installation message returned when installing from URL or local zip file as failure to install; also outputs any installation message now in the console; now has flags forfail_fast
mode (first package that fails stops installation process) and whether to returndetails
(dict per package rather than just a bool for all packages) - method
install_package
(module:weka.core.packages
) can return either a boolean flag of success or detailed information
Release v0.2.11
- methods
install_packages
andinstall_missing_packages
of moduleweka.core.packages
now allow a list of package names instead of tuples (name, version), assuminglatest
as version - method
get_jclass
in moduleweka.core.classes
can handle primitive classes now as well (egint
->java.lang.Integer.TYPE
) - methods
get_non_public_field
andcall_non_public_method
in moduleweka.core.classes
allow accessing private/protected fields and calling private/protected methods of Java objects, which avoids having to sub-class classes to get public access to them (NB: only works as long as the security manager allows that) - added
split_commandline
method to moduleweka.core.classes
, which splits a command-line into classname and option list tuple - the
Instances
class (module:weka.core.dataset
) now supports slicing - added methods
plot_xmlbif_graph
andxmlbif_to_dot
to moduleweka.plot.graph
for plotting XML BIF graphs generated by BayesNet using GraphViz - added method
plot_graph
to moduleweka.plot.graph
to plot dot or XML BIF graphs
Release v0.2.10
- added
logging_level
parameter to thestart
method of theweka.core.jvm
module, enabling the user to turn off debugging output in an easy way (#40) - added method
cv_splits
to classInstances
from moduleweka.core.dataset
to return a list of train/test tuples as used by cross-validation - the
Tester
class (module:weka.experiments
) now has an option to swap columns/rows for comparing datasets rather than classifiers - the
SimpleExperiment
class and derived classes (module:weka.experiments
) now have the additional parameters in the constructor: class_for_ir_statistics, attribute_id, pred_target_column - the method
is_installed
(module:weka.core.packages
) now can check whether a specific version is installed - added
pww-packages
entry point to allow managing of Weka packges from the command-line (actions: list/info/install/uninstall/suggest/is-installed)
Release v0.2.9
- method
JavaObject.new_instance
in moduleweka.core.classes
now automatically installs packages based on suggestions if the JVM was started with theauto_install
flag enabled. - method
test_model_once
of classEvaluation
(module:weka.classifiers
) now has the additional parameterstore
, which allows the recording of the predictions (necessary for statistics like AUC)
Release v0.2.8
- methods
create_instances_from_lists
andcreate_instances_from_matrices
(moduleweka.core.dataset
) now allow the specification of column names, for input and output variables.
Release v0.2.7
- Added property for attribute indices to
DistanceFunction
class (moduleweka.core.distances
) (thanks to Martin Trat, #39) - improved instantiation of classes, avoiding misleading output of exceptions
JavaArray
class (module:weka.core.classes
) now has__str__
and__repr__
methods that output classname and size