<label> <index1>:<value1> <index2>:<value2> ...

11 1:0.2 2:0.4
2-1 1:0.8 2:0.1
31 1:0.3 2:0.6
4-1 1:0.9 2:0.2

1# Download and build
2git clone https://github.com/cjlin1/libsvm.git
3cd libsvm
4make
5
6# Train a model
7./svm-train train.txt model.txt
8
9# Train with options
10./svm-train -s 0 -t 2 -c 10 -g 0.01 train.txt model.txt

1./svm-predict test.txt model.txt output.txt
2# Accuracy = 95% (19/20)
3
4# With probability estimates
5./svm-train -b 1 train.txt model.txt
6./svm-predict -b 1 test.txt model.txt output.txt

./svm-scale -l 0 -u 1 train.txt > train_scaled.txt
./svm-scale -l 0 -u 1 test.txt > test_scaled.txt

1#include "svm.h"
2
3// svm_problem: holds the training data and labels
4struct svm_problem {
5    int l;                    // number of training samples
6    double *y;                // array of labels
7    struct svm_node **x;      // array of feature vectors
8};
9
10// svm_node: a single feature (index:value pair)
11struct svm_node {
12    int index;    // feature index (1-based, -1 terminates)
13    double value; // feature value
14};
15
16// svm_parameter: contains settings for SVM type, kernel type, and hyperparameters
17struct svm_parameter {
18    int svm_type;     // C_SVC, NU_SVC, ONE_CLASS, EPSILON_SVR, NU_SVR
19    int kernel_type;  // LINEAR, POLY, RBF, SIGMOID
20    double C;         // cost parameter
21    double gamma;     // kernel parameter
22    // ... more fields
23};

1#include <stdio.h>
2#include <stdlib.h>
3#include "svm.h"
4
5int main() {
6    // Define training data: 4 samples, 2 features each
7    struct svm_problem prob;
8    prob.l = 4;
9
10    // Labels
11    double labels[] = {1, -1, 1, -1};
12    prob.y = labels;
13
14    // Feature vectors (each terminated by index=-1)
15    struct svm_node nodes[4][3] = {
16        {{1, 0.2}, {2, 0.4}, {-1, 0}},
17        {{1, 0.8}, {2, 0.1}, {-1, 0}},
18        {{1, 0.3}, {2, 0.6}, {-1, 0}},
19        {{1, 0.9}, {2, 0.2}, {-1, 0}}
20    };
21
22    struct svm_node *x[4];
23    for (int i = 0; i < 4; i++)
24        x[i] = nodes[i];
25    prob.x = x;
26
27    // Set parameters
28    struct svm_parameter param;
29    param.svm_type = C_SVC;
30    param.kernel_type = RBF;
31    param.C = 10;
32    param.gamma = 0.5;
33    param.cache_size = 100;
34    param.eps = 1e-3;
35    param.shrinking = 1;
36    param.probability = 0;
37    param.nr_weight = 0;
38
39    // Check parameters
40    const char *error = svm_check_parameter(&prob, &param);
41    if (error) {
42        printf("Parameter error: %s\n", error);
43        return 1;
44    }
45
46    // Train the model
47    struct svm_model *model = svm_train(&prob, &param);
48
49    // Predict a new sample
50    struct svm_node test[] = {{1, 0.25}, {2, 0.5}, {-1, 0}};
51    double prediction = svm_predict(model, test);
52    printf("Prediction: %.0f\n", prediction);
53
54    // Save and load model
55    svm_save_model("model.txt", model);
56    struct svm_model *loaded = svm_load_model("model.txt");
57
58    // Clean up
59    svm_free_and_destroy_model(&model);
60    svm_free_and_destroy_model(&loaded);
61
62    return 0;
63}

1# Compile with libsvm
2gcc -o my_svm my_svm.c svm.cpp -lstdc++ -lm
3
4# Or with g++
5g++ -o my_svm my_svm.c svm.cpp -lm

1double *target = malloc(prob.l * sizeof(double));
2svm_cross_validation(&prob, &param, 5, target);  // 5-fold CV
3
4int correct = 0;
5for (int i = 0; i < prob.l; i++)
6    if (target[i] == prob.y[i]) correct++;
7printf("CV Accuracy: %.2f%%\n", 100.0 * correct / prob.l);
8free(target);