Risk Area SVM

Consider reading our published paper before playing with the code.

Getting Started

Requirements

• g++
• make
• python 2.7x
• libsqlite3

Installation

The only software that needs to be compiled is LIBSVM.

1. Go to /path/to/rasvm/lib/libsvm-ext/ directory, and type make.

How to use

1. Put your data inside /path/to/rasvm/data/ directory.
2. Adjusts the global parameters inside the file rasvm.py to fit yours needs.
3. Go to /path/to/rasvm/ directory, and run the Risk Area SVM with:
   • For a custom split: python rasvm.py data/dataset/training.data data/dataset/testing.data.
   • For a random 5x2cv: python rasvm.py data/dataset/all.data.
4. The results will be saved under /path/to/rasvm/results/ directory.

Observations

1. Your data needs to be in the LIBSVM format. You can use this to convert CSV data to LIBSVM format.
2. Your can save time by properly setting the global parameter nr_local_worker parameter inside the file rasvm.py. This sets the number of processes to be executed in parallel.
3. The positive class must be denoted by the label 1. The other labels will be automatically converted to -1 and considered as negative.
4. The data is automatically scaled to [-1, 1].