Objective functions for our MLP individuals. More...
#include <MlpObjectives.hpp>
Public Member Functions | |
MlpObjectives (const string &defstr, size_t nclasses) | |
Create from a definition string. | |
virtual size_t | nobj () const |
Get number of objectives; up to the concrete classes to determine how this is done, exactly. | |
virtual string | name (size_t ind) const |
Get a longer descriptive name for print-outs; default is 'objective N' where N is ind+1. | |
MlpObjKind | kinds () const |
Protected Member Functions | |
void | addKindIfInList (const vector< string > &spl, MlpObjKind k, size_t nclasses) |
Add an objective (or multiple class-wise objectives) if the mnemonic is found in the list of strings. | |
Protected Attributes | |
size_t | _nobj |
vector< string > | _objname |
MlpObjKind | _evokinds |
Objective functions for our MLP individuals.
The objectives are always in a specific order: Classwise data fitting measures first, then layerwise measures, and then singular overall measures.
TODO: Should this class be responsible also for evaluating the objectives? It can combine evaluations of the many objectives into a single run through the dataset, whereas a single objective would not know about the others? Then, an MlpIndividual should know its MlpObjectives.. Shouldn't be a problem?
MlpObjectives::MlpObjectives | ( | const string & | defstr, | |
size_t | nclasses | |||
) |
Create from a definition string.
Objective function definitions for our MLP individuals.