Decision Rules

Package implementing decision rules. The package allows user to define and process decision rules as Python objects and perform operations on datasets to which these rules apply.

Three types of problems are supported:

• classification
• regression
• survival

Functionalities includes, but is not limited to:

• serialization and deserialization
• prediction
• summary statistics
• comparison between rules (semantic and syntactic)

Installation

Package can be installed from PyPi:

pip install decision-rules

You can also just clone the repository and install the package locally:

pip install .

Extras

Ruleset factories

Module for transforming rule-based models into decision-rules rulesets.

Usage

Module expose only a single function ruleset_factory, which is used to convert RuleKit-type rulesets into decision-rules rulesets.

Example:

from rulekit.classification import RuleClassifier

from decision_rules.ruleset_factories import ruleset_factory
from decision_rules.classification.ruleset import ClassificationRuleSet

rule_classifier = RuleClassifier()
rule_classifier.fit(X, y)

ruleset: ClassificationRuleSet = ruleset_factory(
    model=rule_classifier,
    X_train=X,
    y_train=y,
)

A parser for MLRules-type rulesets has also been implemented and can be found under decision_rules.ruleset_factories._factories.classification.MLRulesRuleSetFactory. Its .make method takes a list of lines read from MLRules algorithm output file, the dataset and (optionally) a voting metric to calculate rule weights.

Example:

with open("example_MLRules_output.txt") as file:
    ml_rules_lines = file.readlines()
ruleset: ClassificationRuleSet = MLRulesRuleSetFactory().make(
    ml_rules_lines, X_df, y_df, "Precision"
)

Running tests

To run tests of the base package, please run:

python -m unittest discover ./tests/base_tests

To run tests of the extras modules, please run:

python -m unittest discover ./tests/extras

Similarly, you can run tests of the whole package by running:

python -m unittest discover ./tests

Documentation

Full documentation along with some usage examples can be found here.

License

The software is licensed under the MIT License. See the LICENSE file for details.