A class that facilitates training a neural network with an arbitrary optimization algorithm.
|
| GNeuralNetTargetFunction (GNeuralNet &nn, const GMatrix &features, const GMatrix &labels) |
| features and labels should be pre-filtered to contain only continuous values for the neural network. More...
|
|
virtual | ~GNeuralNetTargetFunction () |
|
virtual double | computeError (const GVec &pVector) |
| Copies the vector into the neural network and measures sum-squared error. More...
|
|
virtual void | initVector (GVec &pVector) |
| Copies the neural network weights into the vector. More...
|
|
| GTargetFunction (GRelation *pRelation) |
| Takes ownership of pRelation. More...
|
|
| GTargetFunction (size_t dims) |
|
virtual | ~GTargetFunction () |
|
virtual bool | isStable () |
| This method should return true if computeError is deterministic with respect to the vector being optimized. It should return false if the error depends on some state other than the vector being optimized. This mostly affects whether the optimization algorithms are permitted to remember old error values for efficiency purposes. Stable is assumed, so you should override this method if your target function is not stable. More...
|
|
GRelation * | relation () |
| Returns a (smart) pointer to the relation, which specifies the type (discrete or real) of each element in the vector that is being optimized. More...
|
|