// Implementation of Support vector machine in C++ for linear classifier // By bell, with help from Aaron Eberhardt /* Adapted from Andrej Karpathy's blog', c++ of From "hackers guide to NN", Andrej Karpathy http://karpathy.github.io/neuralnets/ wonderful blog, with concepts and complete javascript code * Implementing "linear classifier neural network" in C++ * * Formula (ax+bx+c=label) vector -> label --------------- [1.2, 0.7] -> +1 [-0.3, 0.5] -> -1 [-3, -1] -> +1 [0.1, 1.0] -> -1 [3.0, 1.1] -> -1 [2.1, -3] -> +1 **/ #include #include #include #include using namespace std; int main() { cout << "Linear classifier" << endl; cout << "Coefficients for the linear expression ax+bx+c=label are learned " "from data" << endl; cout << "vector -> label" << endl; cout << "---------------" << endl; cout << "[1.2, 0.7] -> +1" << endl; cout << "[-0.3, -0.5] -> -1" << endl; cout << "[3, 0.1] -> +1" << endl; cout << "[-0.1, -1.0] -> -1" << endl; cout << "[-1.0, 1.1] -> -1" << endl; cout << "[2.1, -3] -> +1 " << endl; // support vector machine double a = 1, b = -2, c = -1; double step_size = 0.01; // initial parameters double score, x, y, pull, offset; int i, j, label; bool ok = true; srand(time(0)); vector data = {1.2, 0.7, -0.3, -0.5, 3.0, 0.1, -0.1, -1.0, -1.0, 1.1, 2.1, -3.}; vector labels = {1, -1, 1, -1, -1, 1}; // data.push_back(10); cout << fixed << setprecision(2); for (i = 0; i < 800; i++) { // pick each point in turn //j = i % 6; // pick a random data point (data.length); j = rand() % labels.size(); x = data[2 * j]; y = data[2 * j + 1]; label = labels[j]; // compute pull score = a * x + b * y + c; if (i == 770) cout << " printing the last cycles: " << endl; pull = 0.0; if (label == 1 && score < 1) pull = 1; if (label == -1 && score > -1) pull = -1; offset = abs(score - label); if (i < 30 || i > 770) cout << j + 1 << " score " << score << " label " << label << " offset " << offset << " pull " << int(pull) << endl; // compute gradient and update parameters a += step_size * (x * pull - offset * a); // -a is from the regularization b += step_size * (y * pull - offset * b); // -b is from the regularization c += step_size * (1 * pull); } for (i = 0; i < int(labels.size()); i++) { if (((a * data[2 * i] + b * data[2 * i + 1] + c) * labels[i]) < 0.) ok = false; } cout <

simpleSVM