algo::ANN< NET, OPT, MET, O > Class Template Reference

Public Member Functions
	ANN ()
	ANN (const arma::Mat< O > &inputs, const arma::Mat< O > &labels)
	ANN (const arma::Mat< O > &inputs, const arma::Mat< O > &labels, const NET network)
template<class... OptArgs>
	ANN (const arma::Mat< O > &inputs, const arma::Mat< O > &labels, const NET network, bool early=false, const OptArgs &... args)
template<class... OptArgs>
	ANN (const arma::Mat< O > &inputs, const arma::Row< size_t > &labels, const NET network, bool early=false, const OptArgs &... args)
void	Train (const arma::Mat< O > &inputs, const arma::Mat< O > &labels)
template<class... OptArgs>
void	Train (const arma::Mat< O > &inputs, const arma::Mat< O > &labels, const NET network, bool early=false, const OptArgs &... args)
template<class... OptArgs>
void	Train (const arma::Mat< O > &inputs, const arma::Row< size_t > &labels, const NET network, bool early=false, const OptArgs &... args)
void	Train (const arma::Mat< O > &inputs, const arma::Row< size_t > &labels)
void	Predict (const arma::Mat< O > &inputs, arma::Mat< O > &preds)
void	Classify (const arma::Mat< O > &inputs, arma::Row< size_t > &preds)
O	ComputeError (const arma::Mat< O > &inputs, const arma::Mat< O > &labels)
arma::Mat< O >	Parameters ()
template<class Archive >
void	serialize (Archive &ar)

Detailed Description

template<class NET, class OPT = ens::StandardSGD, class MET = mlpack::MeanSquaredError, class O = DTYPE>
class algo::ANN< NET, OPT, MET, O >

Definition at line 22 of file nn.h.

Constructor & Destructor Documentation

ANN() [1/3]

template<class NET , class OPT = ens::StandardSGD, class MET = mlpack::MeanSquaredError, class O = DTYPE>

algo::ANN< NET, OPT, MET, O >::ANN ( )

inline

Non-working model

Definition at line 28 of file nn.h.

{ };

ANN() [2/3]

template<class NET , class OPT , class MET , class O >

algo::ANN< NET, OPT, MET, O >::ANN	(	const arma::Mat< O > &	inputs,
		const arma::Mat< O > &	labels
	)

Parameters

network	: pointer to the network object
args	: optimizer arguments in the given order
inputs	: input data, X
labels	: labels of the input, y

Definition at line 27 of file nn_impl.h.

{
  early_ = false;
  opt_ = std::make_unique<OPT>();
  Train(inputs,labels);
}

ANN() [3/3]

template<class NET , class OPT , class MET , class O >

algo::ANN< NET, OPT, MET, O >::ANN	(	const arma::Mat< O > &	inputs,
		const arma::Mat< O > &	labels,
		const NET	network
	)

Parameters

inputs	: input data, X
labels	: labels of the input, y
network	: network object

Definition at line 36 of file nn_impl.h.

{
  network_ = network;
  early_ = false;
  opt_ = std::make_unique<OPT>();
  Train(inputs,labels);
}

Member Function Documentation

Classify()

template<class NET , class OPT , class MET , class O >

void algo::ANN< NET, OPT, MET, O >::Classify	(	const arma::Mat< O > &	inputs,
		arma::Row< size_t > &	preds
	)

Parameters

inputs	: input data, X
preds	: prediction of labels of the input, \hat{y}

Definition at line 180 of file nn_impl.h.

{
  arma::Mat<O> temp;
  network_.Predict(inputs,temp);
  preds = _OneHotDecode(temp,ulab_);
}

ComputeError()

template<class NET , class OPT , class MET , class O >

O algo::ANN< NET, OPT, MET, O >::ComputeError	(	const arma::Mat< O > &	inputs,
		const arma::Mat< O > &	labels
	)

Parameters

inputs	: input data, X
labels	: labels of the input, y

Definition at line 189 of file nn_impl.h.

{
  arma::Mat<O> preds;
  network_.Predict(inputs,preds);
  return MET::Evaluate(preds, labels)/preds.n_elem;
}

Predict()

template<class NET , class OPT , class MET , class O >

void algo::ANN< NET, OPT, MET, O >::Predict	(	const arma::Mat< O > &	inputs,
		arma::Mat< O > &	preds
	)

Parameters

inputs	: input data, X
preds	: prediction of labels of the input, \hat{y}

Definition at line 173 of file nn_impl.h.

{
  network_.Predict(inputs,preds);
}

Train()

template<class NET , class OPT , class MET , class O >

void algo::ANN< NET, OPT, MET, O >::Train	(	const arma::Mat< O > &	inputs,
		const arma::Mat< O > &	labels
	)

Parameters

inputs	: input data, X
labels	: labels of the input, y

Definition at line 102 of file nn_impl.h.

{
  // Safety Net for learning curve generation from scratch,
  // but you might want to start from a trained model. So future modification
  // might be needed...
  if (network_.Parameters().n_elem != 0)
    network_.Reset();
  if (!early_)
    network_.Train(inputs,labels,*opt_);
  else
  {
    arma::Mat<O> inp,lab,val_inp,val_lab;
 
    mlpack::data::Split(inputs,labels,
                        inp,val_inp,lab,val_lab,0.2);
 
    auto func = [&](const arma::Mat<O>& dummy)
    {
      if (MET::NeedsMinimization)
        return MET::Evaluate(*this,val_inp,val_lab);
      else
        return -MET::Evaluate(*this,val_inp,val_lab);
    };
 
    ens::EarlyStopAtMinLossType<arma::Mat<O>> stop(func,5);
    network_.Train(inp,lab,*opt_,stop);
 
  }
}

---

The documentation for this class was generated from the following files:

• nn.h
• nn_impl.h