The command line tool needs to be installed separately in a location that is recognized by the system or in the PATH. The command line tool can be installed on windows (binaries available), linux and OS X (needs compilation). Installation instructions are provided on Peng's Website.
The module requires data within a vector attribute table to be arranged in a specific order. The classification variable (i.e., class labels) need to be in the first column, except for the cat attribute which is not exported. The class label also needs to be in numerical form, i.e., 1, 2, 3.... rather than 'forest' or 'urban'. Also, the attribute table should not contain any missing values because this causes an erroneous mRMR result.
The algorithm outputs a tab-separated list of attributes, ranked by the most important feature first. The method parameter allows a choice between the Maximum Information Difference (MID) and Mutual Information Quotient (MIQ) feature evaluation criteria, which respectively represent the relevancy and redundancy of the features. The algorithm also shows the ranking of the features based on the conventional maximum relevance method. Additional user options include nfeatures which specifies the number of features that you want to select; nsamples limits the maximum number of samples to base the feature selection, and maxvar limits the maximum number of attributes, both of which can therefore reduce the computation for very large datasets. threshold is the discretization threshold to apply to the continuous variable data, i.e., mean +/- threshold * standard deviation. layer is the attribute layer to be used in the feature selection process.
v.mrmr.py vector=vector_layer layer=1 thres=1.0 nfeatures=50 \ nsamples=10000 maxvar=10000 method=MID
Latest change: Thursday Feb 03 09:32:35 2022 in commit: f17c792f5de56c64ecfbe63ec315307872cf9d5c
© 2003-2022 GRASS Development Team, GRASS GIS 8.2.0 Reference Manual