Weka execution helper

Weka has a highly sophisticated command line usage. This wrapper simplifies to run weka from command line

Basic usage

• Run from command line, use --help to get all options

  python run_weka.py --train-file train.arff --test-file test.arff --model-file model.pmml

• IMPORTANT per default, weka is not called, unless explicitly specified with --run option

  python run_weka.py <options> --run
  
  python run_weka.py --settings-input-file pig.json --run

A worked example

1. Install weka and locate the weka.jar file. This path needs be specified with argument --weka-jar

2. Train classifier zeroR with attribute configuration auto on training set my_train.arff (also used for testing) and save trained model in saved_model.pmml

   python run_weka.py -e train -c zeroR --attribute-config=auto --train-file my_train.arff --test-file my_train.arff --model-file saved_model.pmml --run

3. Predict the samples in my_predict.arff using the previously saved model saved_model.pmml and store the prediction output in prediction.csv

   python run_weka.py -e predict --test-file my_predict.arff --model-file saved_model.pmml --prediction-output-file prediction.csv --run

Settings handling

• Settings can be saved with

  python run_weka.py --settings-output-file settings.json

• Settings can be loaded with

  python run_weka.py --settings-input-file settings.json

• It is also possible, to overwrite settings from file on command line, e.g. change sub-sample-seed to 10, independent from value specified in file

  1. Defaults defined in DEFAULT_SETTINGS in source code (lowest precedence)
  2. Settings loaded from settings file
  3. Settings specified on command line (highest precedence)

  python run_weka.py --settings-input-file settings.json --sub-sample-seed 10

• Settings file rewriting

  python run_weka.py --settings-input-file in.json --settings-output-file out.json --classifier j48