Tools

Landscaping and Parameter Tuning

mlpy includes executable scripts to be used off-the-shelf for landscaping and parameter tuning tasks. The classification and optionally feature ranking operations are organized in a sampling procedure (k-fold or Monte Carlo cross validation).

• svm-landscape: landscaping and regularization parameter (C) tuning
• fda-landscape: landscaping and regularization parameter (C) tuning
• srda-landscape: landscaping and alpha parameter (alpha) tuning
• pda-landscape: landscaping and number of regressions parameter (Nreg) tuning
• dlda-landscape
• nn-landscape: landscaping

Error (mlpy.err()), Matthews Correlation Coefficient (mlpy.mcc()) and optionally Canberra Distance (mlpy.canberra()) are retrieved at each parameter step.

mlpy includes executable scripts to be used exclusively for parameter tuning tasks:

• irelief-sigma: kernel width parameter (sigma) tuning

In order to print help message:

$ command --help

Other Tools

borda

Compute Borda Count, Extraction Indicator, Mean Position Indicator from a text file containing feature lists.

canberra

Compute mean Canberra distance indicator on top-k sublists from a text file containing feature lists and one contanining the top-k positions.

In order to print help message:

$ command --help

The Feature Lists File

The feature lists file is a plain text TAB-separated file where each row is a feature ranking (a feature list).

Example:

feat6 [TAB] feat2 [TAB] ... [TAB] feat1
feat4 [TAB] feat1 [TAB] ... [TAB] feat7
feat4 [TAB] feat9 [TAB] ... [TAB] feat3
feat2 [TAB] feat3 [TAB] ... [TAB] feat9
feat8 [TAB] feat4 [TAB] ... [TAB] feat2