runsvm
Description: Wrapper script to call the svm programs
compute-weights and classify-train.
Usage: runsvm [options] <train data> <train
classes> <test data> <weights> <predictions>
Inputs:
All input and output files are in RDB format.
- train data - the training data, with labels in the first column
- train classes - the training set classifications, labels in the
first column, classes (1 or -1) in the second column
- test data - test data, in the same format as the train data
Outputs:
- weights - a file containing weights for each training set example
- predictions - a file containing class predictions for each test
set example
Options:
- -bin <directory> - Directory in which program binaries
reside.
- -power <value> - Raise the kernel to the given power
(default = 1).
- -coefficient <value> - Multiply the kernel by the given
value (default = 1).
- -constant <value> - Add the given constant to the kernel
(default = 1).
- -diagfactor <value> - Add to the diagonal of the kernel
matrix (default = 0). See SVM software documentation for details.
- -radial - Create a radial basis kernel.
- -normalize - Normalize the kernel.
- -posconstraint <value> - Set an upper bound on the values
of weights for positive training set examples (default = 1).
- -negconstraint <value> - Set an upper bound on the values
of weights for negative training set examples (default = 1).
- -noconstraint - Don't use any constraints on the weights.
- -threshold <value> - Train until the objective function
changes by less than the given value (default = 0.000001).
Calls: compute-weights,
classify-train