CAdaptiveEtaCalculator Class Reference
Adaptive Learning Rate
Calculator Interface class. More...
#include <cgradientfunction.h>
Inheritance diagram for CAdaptiveEtaCalculator:
List of all
members.
Detailed Description
Adaptive Learning Rate Calculator Interface class.
Adaptive Learning Rate (Eta) Calculators calculate the learning
rate for the weights of a gradientfunction. This learning rate can
be calculated for each weight individually, so it gives you more
freedom to update your gradientfunction. For every gradient
function you can specify a eta calculator, before updating the
weights of the gradient function, the eta calculator can modify the
gradient feature list for updating (i.e. multiply the features of
the gradient with the learning rates). This is done in the method
getWeightUpdates(CFeatureList), which has to be implemented by all
subclass. If no eta calculator is defined for a gradient function,
the gradient isn't changed (default for the most gradient
functions). There are 2 different eta calculators at your
disposal:
- CIndividualEtaCalculator:
You can set a constant, but for each weight individual learning
rate. Used for example for Neural Networks.
- CVarioEta:
Calculates the learning rate of a weight according to its standard
deviation. For more details see (Coulom, 1) of the paper
section.
Member Function Documentation
| virtual void
CAdaptiveEtaCalculator::getWeightUpdates
|
( |
CFeatureList *
|
updates |
) |
[pure
virtual] |
|
The documentation for this class was generated from the following
file:
|
