/StrongTree

Primary LanguagePython

Strong Optimal Classification Trees

Screenshot

Code for the paper ''Strong Optimal Classification Trees''.

The code will run in python3 ( version 6 and higher) and require Gurobi9.x solver. The required packages are listed in requirements.txt.

Overview

The content of this repository is as follows:

  • Code contains the implementation of approaches FlowOCT and BendersOCT:

    • FlowOCT.py and BendersOCT.py include the formulation of each approach in gurobipy.

    • FlowOCTReplication.py and BendersOCTReplication.py contains the replication code for each approach which is responsible for creating and solving the problem and producing result

    • We have some utility files as follows

      • Tree.py which creates the binary tree
      • logger.py logs the output of console into a text file

  • Results contains the experiments raw results.

  • plots contains the R script for generating figures and tables from the tabular results.

  • DataSets contains all datasets used for the experiments in the paper

How to Run the Code

  • First install the required packages in the requirements.txt
  • For running an instance of FlowOCT (BendersOCT) run the main function in FlowOCTReplication.py (BendersOCTReplication.py) with the following arguments
    • f : name of the dataset
    • d : maximum depth of the tree
    • t : time limit
    • l : value of _lambda
    • i : sample (for shuffling and splitting data)
    • c : 1 if we do calibration; in this case we train our model on 50% of the data otherwise we train on 75% of the data

You can call the main function within a python file as follows (This is what we do in run_exp.py

import FlowOCTReplication
FlowOCTReplication.main(["-f", 'monk1_enc.csv', "-d", 2, "-t", 3600, "-l", 0, "-i", 1, "-c", 1])

or you could run it from terminal

python3 FlowOCTReplication.py -f monk1_enc.csv -d 2 -t 3600 -l 0 -i 1 -c 1

The result of each instance would get stored in a csv file in ./Results/.

Remember that in the replication files we assume that the name of the class label column is target. If you want to use a dataset of your own choice, change the hard-coded label name to the desired name.