You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
dlib/examples/svm_sparse_ex.cpp

121 lines
4.3 KiB

// The contents of this file are in the public domain. See LICENSE_FOR_EXAMPLE_PROGRAMS.txt
/*
This is an example showing how to use sparse feature vectors with
the dlib C++ library's machine learning tools.
This example creates a simple binary classification problem and shows
you how to train a support vector machine on that data.
The data used in this example will be 100 dimensional data and will
come from a simple linearly separable distribution.
*/
#include <iostream>
#include <ctime>
#include <vector>
#include <dlib/svm.h>
using namespace std;
using namespace dlib;
int main()
{
// In this example program we will be dealing with feature vectors that are sparse (i.e. most
// of the values in each vector are zero). So rather than using a dlib::matrix we can use
// one of the containers from the STL to represent our sample vectors. In particular, we
// can use the std::map to represent sparse vectors. (Note that you don't have to use std::map.
// Any STL container of std::pair objects that is sorted can be used. So for example, you could
// use a std::vector<std::pair<unsigned long,double> > here so long as you took care to sort every vector)
typedef std::map<unsigned long,double> sample_type;
// This is a typedef for the type of kernel we are going to use in this example.
// Since our data is linearly separable I picked the linear kernel. Note that if you
// are using a sparse vector representation like std::map then you have to use a kernel
// meant to be used with that kind of data type.
typedef sparse_linear_kernel<sample_type> kernel_type;
// Here we create an instance of the pegasos svm trainer object we will be using.
svm_pegasos<kernel_type> trainer;
// Here we setup a parameter to this object. See the dlib documentation for a
// description of what this parameter does.
trainer.set_lambda(0.00001);
// Let's also use the svm trainer specially optimized for the linear_kernel and
// sparse_linear_kernel.
svm_c_linear_trainer<kernel_type> linear_trainer;
// This trainer solves the "C" formulation of the SVM. See the documentation for
// details.
linear_trainer.set_c(10);
std::vector<sample_type> samples;
std::vector<double> labels;
// make an instance of a sample vector so we can use it below
sample_type sample;
// Now let's go into a loop and randomly generate 10000 samples.
srand(time(0));
double label = +1;
for (int i = 0; i < 10000; ++i)
{
// flip this flag
label *= -1;
sample.clear();
// now make a random sparse sample with at most 10 non-zero elements
for (int j = 0; j < 10; ++j)
{
int idx = std::rand()%100;
double value = static_cast<double>(std::rand())/RAND_MAX;
sample[idx] = label*value;
}
// let the svm_pegasos learn about this sample.
trainer.train(sample,label);
// Also save the samples we are generating so we can let the svm_c_linear_trainer
// learn from them below.
samples.push_back(sample);
labels.push_back(label);
}
// In addition to the rule we learned with the pegasos trainer, let's also use our
// linear_trainer to learn a decision rule.
decision_function<kernel_type> df = linear_trainer.train(samples, labels);
// Now we have trained our SVMs. Let's test them out a bit.
// Each of these statements prints the output of the SVMs given a particular sample.
// Each SVM outputs a number > 0 if a sample is predicted to be in the +1 class and < 0
// if a sample is predicted to be in the -1 class.
sample.clear();
sample[4] = 0.3;
sample[10] = 0.9;
cout << "This is a +1 example, its SVM output is: " << trainer(sample) << endl;
cout << "df: " << df(sample) << endl;
sample.clear();
sample[83] = -0.3;
sample[26] = -0.9;
sample[58] = -0.7;
cout << "This is a -1 example, its SVM output is: " << trainer(sample) << endl;
cout << "df: " << df(sample) << endl;
sample.clear();
sample[0] = -0.2;
sample[9] = -0.8;
cout << "This is a -1 example, its SVM output is: " << trainer(sample) << endl;
cout << "df: " << df(sample) << endl;
}