SYNOPSIS
svm_learn [options] training-file model-fileDESCRIPTION
TinySVM - tiny SVM package Copyright © 2000-2002 Taku Kudo, All rights reserved.Solver Type:
- -l, --solver-type=INT
- select type of solver. TYPE: 0 - C-SVM (default)
- 1 - C-SVR 2 - One-Class-SVM (experimental)
Kernel Parameter:
- -t, --kernel-type=INT
- select type of kernel function. TYPE: 0 - linear (w * x) (default)
- 1 - polynomial
- (s w * x + r)^d
- 2 - neural
- tanh (s w * x + r)
- 3 - RBF
- exp (-s * ||w-x||^2)
- 4 - ANOVA
- (sum_i [exp(-s * ||w_i-x_i||^2)])^d
- -d, --kernel-degree=INT
- set INT for parameter d in polynomial kernel. (default 1)
- -r, --kernel-param-r=FLOAT
- set FLOAT for parameter r in polynomial kernel. (default 1)
- -s, --kernel-param-s=FLOAT
- set FLOAT for parameter s in polynomial kernel. (default 1)
Optimization Parameter:
- -m, --cache-size=FLOAT
- set FLOAT for cache memory size (MB). (default 40.0)
- -c, --cost=FLOAT
- set FLOAT for cost C of constraints violation, trade-off between training error and margin. (default 1.0)
- -e, --termination-criterion=FLOAT
- set FLOAT for tolerance of termination criterion. (default 0.001)
- -H, --shrinking-size=INT
- set INT for number of iterations variable needs to be optimal before considered for shrinking. (default 100)
- -p, --shrinking-eps=FLOAT
- set FLOAT for initial threshold value of shrinking process. (default 2.0)
- -f, --do-final-check=INT
- do final optimality check for variables removed by shrinking. (default 1)
- -i, --insensitive-loss=FLOAT
- set FLOAT for epsilon in epsilon-insensitive loss function used in C-SVR cost evaluation. (default 0.1)
Miscellaneous:
- -M, --model=FILE
- set FILE, FILE.idx for initial condition model file.
- -I, --sv-index
- write all alpha and gradient to MODEL.idx.
- -W, --compress
- calculate vector w (w * x + b), instead of alpha.
- -V, --verbose
- set verbose mode.
- -v, --version
- show the version of TinySVM and exit.
- -h, --help
- show this help and exit.
TinySVM - tiny SVM package Copyright © 2000-2002 Taku Kudo, All rights reserved.