pKNN + Active Learning README File
==================================

1. Requirements and Installation

Matlab v 7.0 or higher with optimization tool-box is required. For installation, type Install_mex on a Matlab command shell.

2. Quick-start

To test the code, we have provided example_script_pknn_al.m, which runs pKNN classifier with active learning over iris dataset whose kernel and labels are provided in K_iris.mat and lbl_iris.mat

3. Using the code

Following are the main files in the package:

pknnTrain: Trains a pknn classifier
pknnPredict: Predicts class labels for the provided test poitns using a trained pknn classifier model
pknn_active: Actively selects examples to be labeled using the framework of Jain and Kapoor, CVPR'09

Assumptions:
1) The kernel should be positive semi-definite and should have reasonably large rank
2) Labels are assumed to be from 1:numclass (numclass is the number of classes)
3) Code typically performs best in the setting where the number of classes is large and the number of examples per class is reasonably small (Note that this the setting where active learning is most crucial and practical).


4. Citation

Finally, please acknowledge the use of our code with a citation:

Prateek Jain and Ashish Kapoor. "Active Learning for Large Multi-class Problems." Proc. IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), 2009.