ffsvm 0.12.0

A libSVM compatible support vector machine, but up to 10x faster, for games or VR.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38

#!/usr/bin/python

import random
import subprocess

svm_train = "/usr/local/bin/svm-train"
svm_predict = "/usr/local/bin/svm-predict"
problemfile = "problem.in"

svm_types = {"csvm": "0", "nusvm": "1", "e_svr": "3", "nu_svr": "4"}
kernel_types = {"linear": "0", "poly": "1", "rbf": "2", "sigmoid": "3"}
probabilities = {"_prob": "1", "": "0"}

CLASSES = 8
SAMPLES_PER_CLASS = 4
ATTRIBUTES = 8


def produce_models(path):
    for svm_type in svm_types.keys():
        for kernel_type in kernel_types.keys():
            for probablity in probabilities.keys():
                s = svm_types[svm_type]
                t = kernel_types[kernel_type]
                b = probabilities[probablity]

                modelfile = f"{path}/m_{svm_type}_{kernel_type}{probablity}.libsvm"
                predictionfile = f"{modelfile}-predicted"

                subprocess.run([svm_train, "-s", s, "-t",
                                t, "-b", b, f"{path}/{problemfile}", modelfile])

                subprocess.run(
                    [svm_predict, "-b", b, f"{path}/{problemfile}", modelfile, predictionfile])


produce_models("data_sparse")
produce_models("data_dense")