public class NystromKernel<T> extends Kernel<T>
A fast active learning algorithm analog to that used in:
Fast Approximation of Distance Between Elastic Curves using
Kernels
Hedi Tabia; David Picard; Hamid Laga; Philippe-Henri Gosselin
British Machine Vision Conference, Sep 2013, United Kingdom. British Machine
Vision Conference
Constructor and Description |
---|
NystromKernel(Kernel<T> kernel)
Default constructor with kernel to approximate as argument
|
Modifier and Type | Method and Description |
---|---|
void |
activeTrain(java.util.List<TrainingSample<T>> list,
int iterations,
int samples,
int pool) |
java.util.List<TrainingSample<double[]>> |
projectList(java.util.List<TrainingSample<T>> l) |
double[] |
projectSample(T sample)
Project a sample to the space induced by the Nystrom approx
|
void |
train(java.util.List<TrainingSample<T>> list)
Train the Nystrom approx on a full training set.
|
double |
valueOf(T t1)
kernel similarity to zero
|
double |
valueOf(T t1,
T t2)
compute the kernel similarity between two element of input space
|
getKernelMatrix, getNormalizedKernelMatrix, normalizedValueOf, setName, toString
public void train(java.util.List<TrainingSample<T>> list)
Might be costly for large training sets, since it involves the inversion of the Gram matrix.
list
- the training list of samplespublic void activeTrain(java.util.List<TrainingSample<T>> list, int iterations, int samples, int pool)
public double[] projectSample(T sample)
sample
- public java.util.List<TrainingSample<double[]>> projectList(java.util.List<TrainingSample<T>> l)
public double valueOf(T t1, T t2)
Kernel