Implementation Task: a feature selection application

Pointwise Mutual Information and Chi-Squared algorithms implementation for feature selection for binary features.

Uses Grand Central Dispatch for multithreading. If it's available in your system it should kick in automatically

How to use

Dist directory contains binary file compiled for x64 Mac OS X. Binary accepts 4 parameters:

1. File from which to read the dataset (defaults to sample.dat). File should be in LIBSVM format

2. Feature selection metric — "pmi" or "chi2" (defaults to chi2)

3. Number of features to select (defaults to 10)

4. File in which to store features (defaults to result.txt)

Here are some examples:

./ml trivial.dat chi2 25 result1.txt
./ml a9a.dat pmi 25 result1.txt

How to compile

Just open ml.xcodeproj in XCode and build.

There is also GNUStep makefile available for Linux compilation, but this option is not currently supported due to author's excessive usage of Objective-C 2.0's syntactic sugar. Please forgive me.