2009-02-17 09:45:57 +08:00
|
|
|
// The contents of this file are in the public domain. See LICENSE_FOR_EXAMPLE_PROGRAMS.txt
|
2008-07-06 00:26:22 +08:00
|
|
|
/*
|
|
|
|
|
|
|
|
This is an example illustrating the use of the rank_features() function
|
|
|
|
from the dlib C++ Library.
|
|
|
|
|
|
|
|
This example creates a simple set of data and then shows
|
|
|
|
you how to use the rank_features() function to find a good
|
|
|
|
set of features (where "good" means the feature set will probably
|
|
|
|
work well with a classification algorithm).
|
|
|
|
|
|
|
|
The data used in this example will be 4 dimensional data and will
|
|
|
|
come from a distribution where points with a distance less than 10
|
|
|
|
from the origin are labeled +1 and all other points are labeled
|
|
|
|
as -1. Note that this data is conceptually 2 dimensional but we
|
|
|
|
will add two extra features for the purpose of showing what
|
|
|
|
the rank_features() function does.
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
|
|
#include <iostream>
|
|
|
|
#include "dlib/svm.h"
|
|
|
|
#include "dlib/rand.h"
|
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
using namespace std;
|
|
|
|
using namespace dlib;
|
|
|
|
|
|
|
|
|
|
|
|
int main()
|
|
|
|
{
|
|
|
|
|
|
|
|
// This first typedef declares a matrix with 4 rows and 1 column. It will be the
|
|
|
|
// object that contains each of our 4 dimensional samples.
|
|
|
|
typedef matrix<double, 4, 1> sample_type;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
// Now lets make some vector objects that can hold our samples
|
|
|
|
std::vector<sample_type> samples;
|
|
|
|
std::vector<double> labels;
|
|
|
|
|
|
|
|
dlib::rand::float_1a rnd;
|
|
|
|
|
2008-10-07 10:24:23 +08:00
|
|
|
for (int x = -30; x <= 30; ++x)
|
2008-07-06 00:26:22 +08:00
|
|
|
{
|
2008-10-07 10:24:23 +08:00
|
|
|
for (int y = -30; y <= 30; ++y)
|
2008-07-06 00:26:22 +08:00
|
|
|
{
|
|
|
|
sample_type samp;
|
|
|
|
|
|
|
|
// the first two features are just the (x,y) position of our points and so
|
|
|
|
// we expect them to be good features since our two classes here are points
|
|
|
|
// close to the origin and points far away from the origin.
|
|
|
|
samp(0) = x;
|
|
|
|
samp(1) = y;
|
|
|
|
|
|
|
|
// This is a worthless feature since it is just random noise. It should
|
|
|
|
// be indicated as worthless by the rank_features() function below.
|
|
|
|
samp(2) = rnd.get_random_double();
|
|
|
|
|
|
|
|
// This is a version of the y feature that is corrupted by random noise. It
|
|
|
|
// should be ranked as less useful than features 0, and 1, but more useful
|
|
|
|
// than the above feature.
|
2008-10-07 10:24:23 +08:00
|
|
|
samp(3) = y*0.2 + (rnd.get_random_double()-0.5)*10;
|
2008-07-06 00:26:22 +08:00
|
|
|
|
|
|
|
// add this sample into our vector of samples.
|
|
|
|
samples.push_back(samp);
|
|
|
|
|
2008-10-07 10:24:23 +08:00
|
|
|
// if this point is less than 15 from the origin then label it as a +1 class point.
|
2008-07-06 00:26:22 +08:00
|
|
|
// otherwise it is a -1 class point
|
2008-10-07 10:24:23 +08:00
|
|
|
if (sqrt((double)x*x + y*y) <= 15)
|
2008-07-06 00:26:22 +08:00
|
|
|
labels.push_back(+1);
|
|
|
|
else
|
|
|
|
labels.push_back(-1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2008-07-07 01:26:32 +08:00
|
|
|
// Here we normalize all the samples by subtracting their mean and dividing by their standard deviation.
|
2008-07-06 00:26:22 +08:00
|
|
|
// This is generally a good idea since it often heads off numerical stability problems and also
|
|
|
|
// prevents one large feature from smothering others.
|
|
|
|
const sample_type m(mean(vector_to_matrix(samples))); // compute a mean vector
|
|
|
|
const sample_type sd(reciprocal(sqrt(variance(vector_to_matrix(samples))))); // compute a standard deviation vector
|
|
|
|
// now normalize each sample
|
|
|
|
for (unsigned long i = 0; i < samples.size(); ++i)
|
|
|
|
samples[i] = pointwise_multiply(samples[i] - m, sd);
|
|
|
|
|
|
|
|
// This is another thing that is often good to do from a numerical stability point of view.
|
2008-07-09 07:39:22 +08:00
|
|
|
// However, in our case it doesn't really matter.
|
2008-07-06 00:26:22 +08:00
|
|
|
randomize_samples(samples,labels);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
// This is a typedef for the type of kernel we are going to use in this example.
|
|
|
|
// In this case I have selected the radial basis kernel that can operate on our
|
|
|
|
// 4D sample_type objects. In general, I would suggest using the same kernel for
|
|
|
|
// classification and feature ranking.
|
|
|
|
typedef radial_basis_kernel<sample_type> kernel_type;
|
|
|
|
|
2009-03-16 10:58:47 +08:00
|
|
|
// Here we declare an instance of the kcentroid object. It is used by rank_features()
|
2009-03-16 11:26:54 +08:00
|
|
|
// two represent the centroids of the two classes. The kcentroid has 3 parameters
|
2009-03-16 10:58:47 +08:00
|
|
|
// you need to set. The first argument to the constructor is the kernel we wish to
|
|
|
|
// use. The second is a parameter that determines the numerical accuracy with which
|
|
|
|
// the object will perform part of the ranking algorithm. Generally, smaller values
|
|
|
|
// give better results but cause the algorithm to attempt to use more support vectors
|
|
|
|
// (and thus run slower and use more memory). The third argument, however, is the
|
|
|
|
// maximum number of support vectors a kcentroid is allowed to use. So you can use
|
2009-03-16 11:26:54 +08:00
|
|
|
// it to control the runtime complexity.
|
|
|
|
kcentroid<kernel_type> kc(kernel_type(0.05), 0.001, 25);
|
2008-07-06 00:26:22 +08:00
|
|
|
|
|
|
|
// And finally we get to the feature ranking. Here we call rank_features() with the kcentroid we just made,
|
2008-07-06 01:32:47 +08:00
|
|
|
// the samples and labels we made above, and the number of features we want it to rank.
|
2009-03-16 10:58:47 +08:00
|
|
|
cout << rank_features(kc, samples, labels) << endl;
|
2008-07-06 00:26:22 +08:00
|
|
|
|
|
|
|
// The output is:
|
|
|
|
/*
|
2009-03-16 10:58:47 +08:00
|
|
|
1 0.514254
|
|
|
|
0 0.810668
|
|
|
|
3 1
|
|
|
|
2 0.994169
|
2008-07-06 00:26:22 +08:00
|
|
|
*/
|
|
|
|
|
|
|
|
// The first column is a list of the features in order of decreasing goodness. So the rank_features() function
|
|
|
|
// is telling us that the samples[i](0) and samples[i](1) (i.e. the x and y) features are the best two. Then
|
|
|
|
// after that the next best feature is the samples[i](3) (i.e. the y corrupted by noise) and finally the worst
|
|
|
|
// feature is the one that is just random noise. So in this case rank_features did exactly what we would
|
|
|
|
// intuitively expect.
|
|
|
|
|
|
|
|
|
2008-10-07 10:24:23 +08:00
|
|
|
// The second column of the matrix is a number that indicates how much the features up to that point
|
|
|
|
// contribute to the separation of the two classes. So bigger numbers are better since they
|
|
|
|
// indicate a larger separation.
|
2008-07-06 00:26:22 +08:00
|
|
|
|
|
|
|
// So to break it down a little more.
|
2009-03-16 10:58:47 +08:00
|
|
|
// 1 0.514254 <-- class separation of feature 1 all by itself
|
|
|
|
// 0 0.810668 <-- class separation of feature 1 and 0
|
2008-10-07 10:24:23 +08:00
|
|
|
// 3 1 <-- class separation of feature 1, 0, and 3
|
2009-03-16 10:58:47 +08:00
|
|
|
// 2 0.994169 <-- class separation of feature 1, 0, 3, and 2
|
2008-10-07 10:24:23 +08:00
|
|
|
|
2008-07-06 00:26:22 +08:00
|
|
|
|
|
|
|
}
|
|
|
|
|