Learns a Random Forest (RF) classifier from data using WEKA, and then converts it into a Sentential Decision Diagram (SDD).

The input is specified using "config.json". The fields that should be configured are:

• "dataset": the name of the dataset
• "tree_count": the maximum number of trees for the RF (passed to WEKA)
• "tree_depth": the maximum depth for each tree in the RF (passed to WEKA)
• "threshold_ratio": the ratio for merging similar thresholds; a ratio of 1.0 does not do any merging

The training and testing file of the dataset should be located at "data/%s/%s.train.csv" and "data/%s/%s.test.csv" after replacing each occurrence of %s with config["dataset"].

For example if config["dataset"] = "magic" then the training and testing file should be located at "data/magic/magic.train.csv" and "data/magic/magic.test.csv"

The output of the program will be 4 files:

• SDD file: The SDD representation of the decision function of the learned RF classifier
• vtree file: The vtree accompanying the SDD
• variable description file: A description of the variables of the SDD (includes variable order, metadata)

After replacing each occurrence of %s with config["dataset"], the SDD file will be written at config["sdd_filename"], the vtree file will be written at config["vtree_filename"], and the variable description file will be written at config["constraint_filename_output"].

./run

Contact us at

Andy Shih: andyshih@cs.ucla.edu
Arthur Choi: aychoi@cs.ucla.edu
Adnan Darwiche: darwiche@cs.ucla.edu