Updated to work with changed ranking stuff.

--HG--
extra : convert_revision : svn%3Afdd8eb12-d10e-0410-9acb-85c331704f74/trunk%403244
This commit is contained in:
Davis King 2009-10-16 01:17:14 +00:00
parent 4fd8980a72
commit a36189a5fc

View File

@ -1,20 +1,20 @@
// The contents of this file are in the public domain. See LICENSE_FOR_EXAMPLE_PROGRAMS.txt
/*
This is an example illustrating the use of the feature ranking
tools from the dlib C++ Library.
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 feature ranking to find a good set of features (where
"good" means the feature set will probably work well with a
classification algorithm).
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
feature ranking does.
the rank_features() function does.
*/
@ -55,7 +55,7 @@ int main()
samp(1) = y;
// This is a worthless feature since it is just random noise. It should
// be indicated as worthless by the feature ranking below.
// 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
@ -85,43 +85,64 @@ int main()
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.
// However, in our case it doesn't matter. It's just here to show you how to do it.
// However, in our case it doesn't really matter. It's just here to show you how to do it.
randomize_samples(samples,labels);
// Finally we get to the feature ranking. Here we call verbose_rank_features_rbf() with
// the samples and labels we made above. The 20 is a measure of how much memory and CPU
// resources the algorithm should use. Generally bigger values give better results but
// take longer to run.
cout << verbose_rank_features_rbf(samples, labels, 20) << endl;
// 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;
// The radial_basis_kernel has a parameter called gamma that we need to set. Generally,
// you should try the same gamma that you are using for training. But if you don't
// have a particular gamma in mind then you can use the following function to
// find a reasonable default gamma for your data.
const double gamma = verbose_find_gamma_with_big_centroid_gap(samples, labels);
// Next we declare an instance of the kcentroid object. It is used by rank_features()
// two represent the centroids of the two classes. The kcentroid has 3 parameters
// 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
// it to put an upper limit on the runtime complexity.
kcentroid<kernel_type> kc(kernel_type(gamma), 0.001, 25);
// And finally we get to the feature ranking. Here we call rank_features() with the kcentroid we just made,
// the samples and labels we made above, and the number of features we want it to rank.
cout << rank_features(kc, samples, labels) << endl;
// The output is:
/*
0 0.810087
0 0.749265
1 1
3 0.873991
2 0.668913
3 0.933378
2 0.825179
*/
// The first column is a list of the features in order of decreasing goodness. So the feature ranking function
// 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 the feature ranking did exactly what we would
// feature is the one that is just random noise. So in this case rank_features did exactly what we would
// intuitively expect.
// 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.
// indicate a larger separation. The max value is always 1. In the case below we see that the bad
// features actually make the class separation go down.
// So to break it down a little more.
// 1 0.810087 <-- class separation of feature 1 all by itself
// 0 1 <-- class separation of feature 1 and 0
// 3 0.873991 <-- class separation of feature 1, 0, and 3
// 2 0.668913 <-- class separation of feature 1, 0, 3, and 2
// 0 0.749265 <-- class separation of feature 0 all by itself
// 1 1 <-- class separation of feature 0 and 1
// 3 0.933378 <-- class separation of feature 0, 1, and 3
// 2 0.825179 <-- class separation of feature 0, 1, 3, and 2
}