/pumil

Convex Formulation of Multiple Instance Learning from Positive and Unlabeled Bags

Primary LanguagePython

Risk Minimization Framework for Multiple Instance Learning from Positive and Unlabeled Bags

Environment

Python 3.4.5 with the following modules:

  • numpy
  • scipy
  • scikit-learn
  • chainer
  • gurobipy

If you do not have Gurobi license, you can substitue cvxopt or openopt for it by rewriting

_SOLVER = 'gurobi'

to each module in MI/PU/SKC.py. Please refer to their official references for further information.

Preparation

In order to prepare datasets, run

./prepare.sh

This script will download datasets from here and make necessary changes.

Experiment

You can make experiments by puskc.py like

python puskc.py --dataset [musk1|musk2|elephant|fox|tiger] --prior [true class prior]