About

A small command-line Java application for analyzing feature impact in decision tree and decision tree ensemble models.

Description of the algorithm:

• WHY did your model predict THAT? (Part 1 of 2)
• WHY did your model predict THAT? (Part 2 of 2)

Installation

Build using Apache Maven:

$ mvn clean package

The build produces an executable uber-JAR file target/feature_impact-executable-1.0-SNAPSHOT.jar.

Development

Initialize Eclipse IDE project using Apache Maven:

$ mvn eclipse:eclipse

Import the project into Eclipse IDE using the menu path File -> Import... -> General/Existing Projects into Workspace (-> Select root directory -> Finish).

The project should be now visible as rf_feature_impact under "Project explorer" and/or "Package explorer" views.

Usage

The resources folder src/main/resources contains LightGBM, XGBoost and Scikit-Learn examples.

Getting help:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --help

Binary classification

Scoring the LightGBM example:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/LightGBMAudit.pmml --target-class 0 --csv-input src/main/resources/csv/Audit.csv --csv-output Audit-impact.csv
$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/LightGBMAudit.pmml --target-class 1 --aggregate true --csv-input src/main/resources/csv/Audit.csv --csv-output Audit-aggregate_impact.csv

Scoring the XGBoost example:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/XGBoostAudit.pmml --target-class 0 --csv-input src/main/resources/csv/Audit.csv --csv-output Audit-impact.csv
$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/XGBoostAudit.pmml --target-class 1 --aggregate true --csv-input src/main/resources/csv/Audit.csv --csv-output Audit-aggregate_impact.csv

Scoring the Scikit-Learn random forest example:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/RandomForestAudit.pmml --target-class 0 --csv-input src/main/resources/csv/Audit.csv --csv-output Audit-impact.csv
$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/RandomForestAudit.pmml --target-class 1 --aggregate true --csv-input src/main/resources/csv/Audit.csv --csv-output Audit-aggregate_impact.csv

Regression

Scoring the LightGBM example:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/LightGBMAuto.pmml --csv-input src/main/resources/csv/Auto.csv --csv-output Auto-impact.csv
$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/LightGBMAuto.pmml --aggregate true --csv-input src/main/resources/csv/Auto.csv --csv-output Auto-aggregate_impact.csv

Scoring the XGBoost example:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/XGBoostAuto.pmml --csv-input src/main/resources/csv/Auto.csv --csv-output Auto-impact.csv
$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/XGBoostAuto.pmml --aggregate true --csv-input src/main/resources/csv/Auto.csv --csv-output Auto-aggregate_impact.csv

Scoring the Scikit-Learn example:

$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/RandomForestAuto.pmml --csv-input src/main/resources/csv/Auto.csv --csv-output Auto-impact.csv
$ java -jar target/feature_impact-executable-1.0-SNAPSHOT.jar --pmml-model src/main/resources/pmml/RandomForestAuto.pmml --aggregate true --csv-input src/main/resources/csv/Auto.csv --csv-output Auto-aggregate_impact.csv