gist-rfe

Description:

Perform recursive feature elimination, following

I. Guyon, J. Weston, S. Barnhill and V. Vapnik. "Gene selection for cancer classification using support vector machines." Machine Learning. 46(1-3):389-422, 2002.

This is an algorithm for selecting a subset of features for a particular learning task. The basic algorithm is the following:

1. Initialize the data set to contain all features.
2. Train an SVM on the data set.
3. Rank features according to c_i = (w_i)².
4. Eliminate the lower-ranked 50% of the features.
5. If more than one feature remains, return to step 2.

When using this algorithm, beware of incurring a selection bias. For details, see

C. Ambroise and G. J. McLachlan. "Selection bias in gene extraction on the basis of microarray gene-expression data." PNAS. 99:6562-6566, 2002.

Usage: gist-rfe [options] <train data> <train labels>

Inputs:

• train data - a data file suitable for input to compute-weights
• train labels - the corresponding label file

Output:

Prints to standard output the performance evaluation (as produced by score-svm-results) at each iteration of the SVM-RFE algorithm. By default, evaluates the performance using leave-one-out cross-validation. The -test option enables evaluation on an independent test set.

Options:

• -test <data> <labels> - Evaluate performance with respect to an external test set.
• -rfe-iter <value> - Perform a maximum number of iterations of RFE. By default, the algorithm continues until all features are eliminated.
• -reduce <percent> - Percent of training set to eliminate each round. At least one feaure will always be removed. Default = 50.
• -features <file> - File in which to print the features selected at each iteration. The output is in four columns: feature name, iteration, score (c_i) and a binary value indicating whether the feature was discarded in this iteration. If the filename contains a %, then multiple output files will be created, replacing % with the iteration number.
• -weights <file> - File in which to store weights. The same use of % applies.
• -predicts <file> - File in which to store predictions. The same use of % applies.

  In addition, any option that is valid for compute-weights may also be given to SVM-RFE.

Calls: compute-weights, classify, score-svm-results