probabilistic_layer.cpp

//   OpenNN: Open Neural Networks Library
//   www.opennn.net
//
//   P R O B A B I L I S T I C   L A Y E R   C L A S S
//
//   Artificial Intelligence Techniques SL
//   artelnics@artelnics.com
 
#include "probabilistic_layer.h"
 
namespace OpenNN
{
 
 
ProbabilisticLayer::ProbabilisticLayer()
{
    set();
}
 
 
 
ProbabilisticLayer::ProbabilisticLayer(const Index& new_inputs_number, const Index& new_neurons_number)
{
    set(new_inputs_number, new_neurons_number);
 
    if(new_neurons_number > 1)
    {
        activation_function = ActivationFunction::Softmax;
    }
}
 
 
 
ProbabilisticLayer::~ProbabilisticLayer()
{
}
 
 
Index ProbabilisticLayer::get_inputs_number() const
{
    return synaptic_weights.dimension(0);
}
 
 
Index ProbabilisticLayer::get_neurons_number() const
{
    return biases.size();
}
 
 
Index ProbabilisticLayer::get_biases_number() const
{
    return biases.size();
}
 
 
 
Index ProbabilisticLayer::get_synaptic_weights_number() const
{
    return synaptic_weights.size();
}
 
 
 
const type& ProbabilisticLayer::get_decision_threshold() const
{
    return decision_threshold;
}
 
 
 
const ProbabilisticLayer::ActivationFunction& ProbabilisticLayer::get_activation_function() const
{
    return activation_function;
}
 
 
 
string ProbabilisticLayer::write_activation_function() const
{
    if(activation_function == ActivationFunction::Binary)
    {
        return "Binary";
    }
    else if(activation_function == ActivationFunction::Logistic)
    {
        return "Logistic";
    }
    else if(activation_function == ActivationFunction::Competitive)
    {
        return "Competitive";
    }
    else if(activation_function == ActivationFunction::Softmax)
    {
        return "Softmax";
    }
    else
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "string write_activation_function() const method.\n"
               << "Unknown probabilistic method.\n";
 
        throw logic_error(buffer.str());
    }
}
 
 
 
string ProbabilisticLayer::write_activation_function_text() const
{
    if(activation_function == ActivationFunction::Binary)
    {
        return "binary";
    }
    else if(activation_function == ActivationFunction::Logistic)
    {
        return "logistic";
    }
    else if(activation_function == ActivationFunction::Competitive)
    {
        return "competitive";
    }
    else if(activation_function == ActivationFunction::Softmax)
    {
        return "softmax";
    }
    else
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "string write_activation_function_text() const method.\n"
               << "Unknown probabilistic method.\n";
 
        throw logic_error(buffer.str());
    }
}
 
 
 
const bool& ProbabilisticLayer::get_display() const
{
    return display;
}
 
 
 
const Tensor<type, 2>& ProbabilisticLayer::get_biases() const
{
    return biases;
}
 
 
 
const Tensor<type, 2>& ProbabilisticLayer::get_synaptic_weights() const
{
    return synaptic_weights;
}
 
 
 
Tensor<type, 2> ProbabilisticLayer::get_biases(Tensor<type, 1>& parameters) const
{
    const Index neurons_number = get_neurons_number();
 
    const TensorMap < Tensor<type, 2> > bias_tensor(parameters.data(),  1, neurons_number);
 
    return bias_tensor;
}
 
 
 
Tensor<type, 2> ProbabilisticLayer::get_synaptic_weights(Tensor<type, 1>& parameters) const
{
    const Index inputs_number = get_inputs_number();
    const Index neurons_number = get_neurons_number();
    const Index biases_number = get_biases_number();
 
    const TensorMap< Tensor<type, 2> > synaptic_weights_tensor(parameters.data()+biases_number, inputs_number, neurons_number);
 
    return  synaptic_weights_tensor;
}
 
 
 
Index ProbabilisticLayer::get_parameters_number() const
{
    return biases.size() + synaptic_weights.size();
}
 
 
 
Tensor<type, 1> ProbabilisticLayer::get_parameters() const
{
    Tensor<type, 1> parameters(synaptic_weights.size() + biases.size());
 
    for(Index i = 0; i < biases.size(); i++)
    {
        fill_n(parameters.data()+i, 1, biases(i));
    }
 
    for(Index i = 0; i < synaptic_weights.size(); i++)
    {
        fill_n(parameters.data()+ biases.size() +i, 1, synaptic_weights(i));
    }
 
    return parameters;
 
}
 
 
 
void ProbabilisticLayer::set()
{
    biases.resize(0, 0);
 
    synaptic_weights.resize(0,0);
 
    set_default();
}
 
 
 
void ProbabilisticLayer::set(const Index& new_inputs_number, const Index& new_neurons_number)
{
    biases.resize(1, new_neurons_number);
 
    synaptic_weights.resize(new_inputs_number, new_neurons_number);
 
    set_parameters_random();
 
    set_default();
}
 
 
 
void ProbabilisticLayer::set(const ProbabilisticLayer& other_probabilistic_layer)
{
    set_default();
 
    activation_function = other_probabilistic_layer.activation_function;
 
    decision_threshold = other_probabilistic_layer.decision_threshold;
 
    display = other_probabilistic_layer.display;
}
 
 
void ProbabilisticLayer::set_inputs_number(const Index& new_inputs_number)
{
    const Index neurons_number = get_neurons_number();
 
    biases.resize(1, neurons_number);
 
    synaptic_weights.resize(new_inputs_number, neurons_number);
}
 
 
void ProbabilisticLayer::set_neurons_number(const Index& new_neurons_number)
{
    const Index inputs_number = get_inputs_number();
 
    biases.resize(1, new_neurons_number);
 
    synaptic_weights.resize(inputs_number, new_neurons_number);
}
 
 
void ProbabilisticLayer::set_biases(const Tensor<type, 2>& new_biases)
{
    biases = new_biases;
}
 
 
void ProbabilisticLayer::set_synaptic_weights(const Tensor<type, 2>& new_synaptic_weights)
{
    synaptic_weights = new_synaptic_weights;
}
 
 
void ProbabilisticLayer::set_parameters(const Tensor<type, 1>& new_parameters, const Index& index)
{
    const Index biases_number = biases.size();
    const Index synaptic_weights_number = synaptic_weights.size();
 
    memcpy(biases.data(), new_parameters.data() + index, static_cast<size_t>(biases_number)*sizeof(type));
    memcpy(synaptic_weights.data(), new_parameters.data() + biases_number + index, static_cast<size_t>(synaptic_weights_number)*sizeof(type));
}
 
 
 
void ProbabilisticLayer::set_decision_threshold(const type& new_decision_threshold)
{
#ifdef OPENNN_DEBUG
 
    if(new_decision_threshold <= 0)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_decision_threshold(const type&) method.\n"
               << "Decision threshold(" << decision_threshold << ") must be greater than zero.\n";
 
        throw logic_error(buffer.str());
    }
    else if(new_decision_threshold >= 1)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_decision_threshold(const type&) method.\n"
               << "Decision threshold(" << decision_threshold << ") must be less than one.\n";
 
        throw logic_error(buffer.str());
    }
 
#endif
 
    decision_threshold = new_decision_threshold;
}
 
 
 
void ProbabilisticLayer::set_default()
{
    layer_name = "probabilistic_layer";
 
    layer_type = Layer::Type::Probabilistic;
 
    const Index neurons_number = get_neurons_number();
 
    if(neurons_number == 1)
    {
        activation_function = ActivationFunction::Logistic;
    }
    else
    {
        activation_function = ActivationFunction::Softmax;
    }
 
    decision_threshold = type(0.5);
 
    display = true;
}
 
 
 
void ProbabilisticLayer::set_activation_function(const ActivationFunction& new_activation_function)
{
#ifdef OPENNN_DEBUG
 
    const Index neurons_number = get_neurons_number();
 
    if(neurons_number == 1 && new_activation_function == Competitive)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_activation_function(const ActivationFunction&) method.\n"
               << "Activation function cannot be Competitive when the number of neurons is 1.\n";
 
        throw logic_error(buffer.str());
    }
 
    if(neurons_number == 1 && new_activation_function == Softmax)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_activation_function(const ActivationFunction&) method.\n"
               << "Activation function cannot be Softmax when the number of neurons is 1.\n";
 
        throw logic_error(buffer.str());
    }
 
    if(neurons_number != 1 && new_activation_function == Binary)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_activation_function(const ActivationFunction&) method.\n"
               << "Activation function cannot be Binary when the number of neurons is greater than 1.\n";
 
        throw logic_error(buffer.str());
    }
 
    if(neurons_number != 1 && new_activation_function == Logistic)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_activation_function(const ActivationFunction&) method.\n"
               << "Activation function cannot be Logistic when the number of neurons is greater than 1.\n";
 
        throw logic_error(buffer.str());
    }
 
#endif
 
    activation_function = new_activation_function;
}
 
 
 
void ProbabilisticLayer::set_activation_function(const string& new_activation_function)
{
    if(new_activation_function == "Binary")
    {
        set_activation_function(ActivationFunction::Binary);
    }
    else if(new_activation_function == "Logistic")
    {
        set_activation_function(ActivationFunction::Logistic);
    }
    else if(new_activation_function == "Competitive")
    {
        set_activation_function(ActivationFunction::Competitive);
    }
    else if(new_activation_function == "Softmax")
    {
        set_activation_function(ActivationFunction::Softmax);
    }
    else
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void set_activation_function(const string&) method.\n"
               << "Unknown probabilistic method: " << new_activation_function << ".\n";
 
        throw logic_error(buffer.str());
    }
}
 
 
 
void ProbabilisticLayer::set_display(const bool& new_display)
{
    display = new_display;
}
 
 
 
void ProbabilisticLayer::set_biases_constant(const type& value)
{
    biases.setConstant(value);
}
 
 
 
void ProbabilisticLayer::set_synaptic_weights_constant(const type& value)
{
    synaptic_weights.setConstant(value);
}
 
 
void ProbabilisticLayer::set_synaptic_weights_constant_Glorot()
{
    synaptic_weights.setRandom();
}
 
 
 
void ProbabilisticLayer::set_parameters_constant(const type& value)
{
    biases.setConstant(value);
 
    synaptic_weights.setConstant(value);
}
 
 
 
void ProbabilisticLayer::set_parameters_random()
{
    const type minimum = type(-0.2);
    const type maximum = type(0.2);
 
    for(Index i = 0; i < biases.size(); i++)
    {
        const type random = static_cast<type>(rand()/(RAND_MAX+1.0));
 
        biases(i) = minimum + (maximum - minimum)*random;
    }
 
    for(Index i = 0; i < synaptic_weights.size(); i++)
    {
        const type random = static_cast<type>(rand()/(RAND_MAX+1.0));
 
        synaptic_weights(i) = minimum + (maximum - minimum)*random;
    }
}
 
 
void ProbabilisticLayer::insert_parameters(const Tensor<type, 1>& parameters, const Index& )
{
    const Index biases_number = get_biases_number();
    const Index synaptic_weights_number = get_synaptic_weights_number();
 
    memcpy(biases.data() , parameters.data(), static_cast<size_t>(biases_number)*sizeof(type));
    memcpy(synaptic_weights.data(), parameters.data() + biases_number, static_cast<size_t>(synaptic_weights_number)*sizeof(type));
}
 
 
void ProbabilisticLayer::calculate_combinations(const Tensor<type, 2>& inputs,
                                                const Tensor<type, 2>& biases,
                                                const Tensor<type, 2>& synaptic_weights,
                                                Tensor<type, 2>& combinations) const
{
    const Index batch_samples_number = inputs.dimension(0);
    const Index biases_number = get_neurons_number();
 
    for(Index i = 0; i < biases_number; i++)
    {
        fill_n(combinations.data()+i*batch_samples_number, batch_samples_number, biases(i));
    }
 
    combinations.device(*thread_pool_device) += inputs.contract(synaptic_weights, A_B);
}
 
 
// Activations
 
void ProbabilisticLayer::calculate_activations(const Tensor<type, 2>& combinations, Tensor<type, 2>& activations) const
{
#ifdef OPENNN_DEBUG
 
    const Index dimensions_number = combinations.rank();
 
    if(dimensions_number != 2)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void calculate_activations(const Tensor<type, 2>&, Tensor<type, 2>&) const method.\n"
               << "Dimensions of combinations (" << dimensions_number << ") must be 2.\n";
 
        throw logic_error(buffer.str());
    }
 
    const Index neurons_number = get_neurons_number();
 
    const Index combinations_columns_number = combinations.dimension(1);
 
    if(combinations_columns_number != neurons_number)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void calculate_activations(const Tensor<type, 2>&, Tensor<type, 2>&) const method.\n"
               << "Number of combinations columns (" << combinations_columns_number << ") must be equal to number of neurons (" << neurons_number << ").\n";
 
        throw logic_error(buffer.str());
    }
 
#endif
 
    switch(activation_function)
    {
    case ActivationFunction::Binary: binary(combinations, activations); return;
 
    case ActivationFunction::Logistic: logistic(combinations, activations); return;
 
    case ActivationFunction::Competitive: competitive(combinations, activations); return;
 
    case ActivationFunction::Softmax: softmax(combinations, activations); return;
    }
 
    ostringstream buffer;
 
    buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
           << "void calculate_activations(const Tensor<type, 2>&, Tensor<type, 2>&) const method.\n"
           << "Unknown probabilistic method.\n";
 
    throw logic_error(buffer.str());
}
 
 
void ProbabilisticLayer::calculate_activations_derivatives(const Tensor<type, 2>& combinations,
                                                           Tensor<type, 2>& activations,
                                                           Tensor<type, 3>& activations_derivatives) const
{
#ifdef OPENNN_DEBUG
 
    const Index neurons_number = get_neurons_number();
 
    const Index combinations_columns_number = combinations.dimension(1);
 
    if(combinations_columns_number != neurons_number)
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void calculate_activations_derivatives(const Tensor<type, 2>&, Tensor<type, 2>&) const method.\n"
               << "Number of combinations columns (" << combinations_columns_number
               << ") must be equal to number of neurons (" << neurons_number << ").\n";
 
        throw logic_error(buffer.str());
    }
 
#endif
 
    switch(activation_function)
    {
    case ActivationFunction::Logistic: logistic_derivatives(combinations, activations, activations_derivatives); return;
 
    case ActivationFunction::Softmax: softmax_derivatives(combinations, activations, activations_derivatives); return;
 
    default: return;
    }
}
 
 
 
Tensor<type, 2> ProbabilisticLayer::calculate_outputs(const Tensor<type, 2>& inputs)
{
    const Index batch_size = inputs.dimension(0);
    const Index outputs_number = get_neurons_number();
 
    Tensor<type, 2> outputs(batch_size, outputs_number);
 
    calculate_combinations(inputs, biases, synaptic_weights, outputs);
 
    calculate_activations(outputs, outputs);
 
    return outputs;
}
 
 
void ProbabilisticLayer::forward_propagate(const Tensor<type, 2>& inputs, LayerForwardPropagation* forward_propagation)
{
    ProbabilisticLayerForwardPropagation* probabilistic_layer_forward_propagation
            = static_cast<ProbabilisticLayerForwardPropagation*>(forward_propagation);
 
    calculate_combinations(inputs, biases, synaptic_weights, probabilistic_layer_forward_propagation->combinations);
 
    calculate_activations_derivatives(probabilistic_layer_forward_propagation->combinations,
                                      probabilistic_layer_forward_propagation->activations,
                                      probabilistic_layer_forward_propagation->activations_derivatives);
}
 
 
void ProbabilisticLayer::forward_propagate(const Tensor<type, 2>& inputs,
                                           Tensor<type, 1> potential_parameters,
                                           LayerForwardPropagation* forward_propagation)
{
    const Index neurons_number = get_neurons_number();
    const Index inputs_number = get_inputs_number();
 
#ifdef OPENNN_DEBUG
 
    if(inputs_number != inputs.dimension(1))
    {
        ostringstream buffer;
 
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void forward_propagate(const Tensor<type, 2>&, Tensor<type, 1>&, ForwardPropagation&) method.\n"
               << "Number of inputs columns (" << inputs.dimension(1) << ") must be equal to number of inputs ("
               << inputs_number << ").\n";
 
        throw logic_error(buffer.str());
    }
 
#endif
 
    ProbabilisticLayerForwardPropagation* probabilistic_layer_forward_propagation
            = static_cast<ProbabilisticLayerForwardPropagation*>(forward_propagation);
 
    const TensorMap<Tensor<type, 2>> potential_biases(potential_parameters.data(), neurons_number, 1);
 
    const TensorMap<Tensor<type, 2>> potential_synaptic_weights(potential_parameters.data()+neurons_number,
                                                                inputs_number, neurons_number);
 
    calculate_combinations(inputs, potential_biases, potential_synaptic_weights, probabilistic_layer_forward_propagation->combinations);
 
    calculate_activations_derivatives(probabilistic_layer_forward_propagation->combinations,
                                      probabilistic_layer_forward_propagation->activations,
                                      probabilistic_layer_forward_propagation->activations_derivatives);
}
 
 
// Gradient methods
 
void ProbabilisticLayer::calculate_error_gradient(const Tensor<type, 2>& inputs,
                                                  LayerForwardPropagation* forward_propagation,
                                                  LayerBackPropagation* back_propagation) const
{
    ProbabilisticLayerForwardPropagation* probabilistic_layer_forward_propagation =
            static_cast<ProbabilisticLayerForwardPropagation*>(forward_propagation);
 
    ProbabilisticLayerBackPropagation* probabilistic_layer_back_propagation =
            static_cast<ProbabilisticLayerBackPropagation*>(back_propagation);
 
    const Index samples_number = inputs.dimension(0);
    const Index neurons_number = get_neurons_number();
 
    if(neurons_number == 1) // Binary gradient
    {
        TensorMap< Tensor<type, 2> > activations_derivatives(probabilistic_layer_forward_propagation->activations_derivatives.data(), samples_number, neurons_number);
 
        probabilistic_layer_back_propagation->biases_derivatives.device(*thread_pool_device) =
                (probabilistic_layer_back_propagation->delta * activations_derivatives).sum(Eigen::array<Index, 1>({0}));
 
        probabilistic_layer_back_propagation->synaptic_weights_derivatives.device(*thread_pool_device) =
                inputs.contract((probabilistic_layer_back_propagation->delta * activations_derivatives), AT_B);
    }
    else // Multiple gradient
    {
        if(activation_function == ActivationFunction::Softmax)
        {
            const Index step = neurons_number * neurons_number;
 
            for(Index i = 0; i < samples_number; i++)
            {
                probabilistic_layer_back_propagation->delta_row = probabilistic_layer_back_propagation->delta.chip(i,0);
 
                TensorMap< Tensor<type, 2> > activations_derivatives_matrix(probabilistic_layer_forward_propagation->activations_derivatives.data() + i*step,
                                                                            neurons_number, neurons_number);
 
                probabilistic_layer_back_propagation->error_combinations_derivatives.chip(i,0) =
                        probabilistic_layer_back_propagation->delta_row.contract(activations_derivatives_matrix, AT_B);
            }
 
            probabilistic_layer_back_propagation->biases_derivatives.device(*thread_pool_device) =
                    (probabilistic_layer_back_propagation->error_combinations_derivatives).sum(Eigen::array<Index, 1>({0}));
 
            probabilistic_layer_back_propagation->synaptic_weights_derivatives.device(*thread_pool_device) =
                    inputs.contract(probabilistic_layer_back_propagation->error_combinations_derivatives, AT_B);
        }
        else
        {
            TensorMap< Tensor<type, 2> > activations_derivatives(probabilistic_layer_forward_propagation->activations_derivatives.data(), samples_number, neurons_number);
 
            probabilistic_layer_back_propagation->biases_derivatives.device(*thread_pool_device) =
                    (probabilistic_layer_back_propagation->delta*activations_derivatives).sum(Eigen::array<Index, 1>({0}));
 
            probabilistic_layer_back_propagation->synaptic_weights_derivatives.device(*thread_pool_device) =
                    inputs.contract((probabilistic_layer_back_propagation->delta*activations_derivatives), AT_B);
        }
    }
}
 
 
void ProbabilisticLayer::insert_gradient(LayerBackPropagation* back_propagation, const Index& index, Tensor<type, 1>& gradient) const
{
    const Index biases_number = get_biases_number();
    const Index synaptic_weights_number = get_synaptic_weights_number();
 
    const ProbabilisticLayerBackPropagation* probabilistic_layer_back_propagation =
            static_cast<ProbabilisticLayerBackPropagation*>(back_propagation);
 
    memcpy(gradient.data() + index,
           probabilistic_layer_back_propagation->biases_derivatives.data(),
           static_cast<size_t>(biases_number)*sizeof(type));
 
    memcpy(gradient.data() + index + biases_number,
           probabilistic_layer_back_propagation->synaptic_weights_derivatives.data(),
           static_cast<size_t>(synaptic_weights_number)*sizeof(type));
}
 
 
void ProbabilisticLayer::calculate_squared_errors_Jacobian_lm(const Tensor<type, 2>& inputs,
                                                              LayerForwardPropagation* forward_propagation,
                                                              LayerBackPropagationLM* back_propagation)
{
    ProbabilisticLayerForwardPropagation* probabilistic_layer_forward_propagation =
            static_cast<ProbabilisticLayerForwardPropagation*>(forward_propagation);
 
    ProbabilisticLayerBackPropagationLM* probabilistic_layer_back_propagation_lm =
            static_cast<ProbabilisticLayerBackPropagationLM*>(back_propagation);
 
    const Index samples_number = inputs.dimension(0);
 
    const Index inputs_number = get_inputs_number();
    const Index neurons_number = get_neurons_number();
 
    probabilistic_layer_back_propagation_lm->squared_errors_Jacobian.setZero();
 
    if(activation_function == ActivationFunction::Softmax)
    {
        Index parameter_index = 0;
 
        for(Index sample = 0; sample < samples_number; sample++)
        {
            parameter_index = 0;
 
            for(Index neuron = 0; neuron < neurons_number; neuron++)
            {
                for(Index input = 0; input <  inputs_number; input++)
                {
                    probabilistic_layer_back_propagation_lm->squared_errors_Jacobian(sample, neurons_number+parameter_index) =
                            probabilistic_layer_back_propagation_lm->error_combinations_derivatives(sample, neuron) *
                            inputs(sample, input);
 
                    parameter_index++;
                }
 
                probabilistic_layer_back_propagation_lm->squared_errors_Jacobian(sample, neuron) =
                        probabilistic_layer_back_propagation_lm->error_combinations_derivatives(sample, neuron);
            }
        }
    }
    else
    {
        Index parameter_index = 0;
 
        for(Index sample = 0; sample < samples_number; sample++)
        {
            parameter_index = 0;
 
            for(Index neuron = 0; neuron < neurons_number; neuron++)
            {
                for(Index input = 0; input <  inputs_number; input++)
                {
                    probabilistic_layer_back_propagation_lm->squared_errors_Jacobian(sample, neurons_number+parameter_index) =
                            probabilistic_layer_back_propagation_lm->delta(sample, neuron) *
                            probabilistic_layer_forward_propagation->activations_derivatives(sample, neuron, 0) *
                            inputs(sample, input);
 
                    parameter_index++;
                }
 
                probabilistic_layer_back_propagation_lm->squared_errors_Jacobian(sample, neuron) =
                        probabilistic_layer_back_propagation_lm->delta(sample, neuron) *
                        probabilistic_layer_forward_propagation->activations_derivatives(sample, neuron, 0);
            }
        }
    }
}
 
 
void ProbabilisticLayer::insert_squared_errors_Jacobian_lm(LayerBackPropagationLM * back_propagation ,
                                                           const Index & index,
                                                           Tensor<type, 2>& squared_errors_Jacobian) const
{
    ProbabilisticLayerBackPropagationLM* probabilistic_layer_back_propagation_lm =
            static_cast<ProbabilisticLayerBackPropagationLM*>(back_propagation);
 
    const Index batch_samples_number = probabilistic_layer_back_propagation_lm->squared_errors_Jacobian.dimension(0);
    const Index layer_parameters_number = get_parameters_number();
 
    memcpy(squared_errors_Jacobian.data() + index,
           probabilistic_layer_back_propagation_lm->squared_errors_Jacobian.data(),
           static_cast<size_t>(layer_parameters_number*batch_samples_number)*sizeof(type));
}
 
 
 
void ProbabilisticLayer::write_XML(tinyxml2::XMLPrinter& file_stream) const
{
    ostringstream buffer;
 
    // Probabilistic layer
 
    file_stream.OpenElement("ProbabilisticLayer");
 
    // Inputs number
 
    file_stream.OpenElement("InputsNumber");
 
    buffer.str("");
    buffer << get_inputs_number();
 
    file_stream.PushText(buffer.str().c_str());
 
    file_stream.CloseElement();
 
    // Neurons number
 
    file_stream.OpenElement("NeuronsNumber");
 
    buffer.str("");
    buffer << get_neurons_number();
 
    file_stream.PushText(buffer.str().c_str());
 
    file_stream.CloseElement();
 
    // Activation function
 
    file_stream.OpenElement("ActivationFunction");
 
    file_stream.PushText(write_activation_function().c_str());
 
    file_stream.CloseElement();
 
    // Parameters
 
    file_stream.OpenElement("Parameters");
 
    buffer.str("");
 
    const Tensor<type, 1> parameters = get_parameters();
    const Index parameters_size = parameters.size();
 
    for(Index i = 0; i < parameters_size; i++)
    {
        buffer << parameters(i);
 
        if(i != (parameters_size-1)) buffer << " ";
    }
 
    file_stream.PushText(buffer.str().c_str());
 
    file_stream.CloseElement();
 
    // Decision threshold
 
    file_stream.OpenElement("DecisionThreshold");
 
    buffer.str("");
    buffer << decision_threshold;
 
    file_stream.PushText(buffer.str().c_str());
 
    file_stream.CloseElement();
 
    // Probabilistic layer (end tag)
 
    file_stream.CloseElement();
}
 
 
 
void ProbabilisticLayer::from_XML(const tinyxml2::XMLDocument& document)
{
    ostringstream buffer;
 
    // Probabilistic layer
 
    const tinyxml2::XMLElement* probabilistic_layer_element = document.FirstChildElement("ProbabilisticLayer");
 
    if(!probabilistic_layer_element)
    {
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void from_XML(const tinyxml2::XMLDocument&) method.\n"
               << "Probabilistic layer element is nullptr.\n";
 
        throw logic_error(buffer.str());
    }
 
    // Inputs number
 
    const tinyxml2::XMLElement* inputs_number_element = probabilistic_layer_element->FirstChildElement("InputsNumber");
 
    if(!inputs_number_element)
    {
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void from_XML(const tinyxml2::XMLDocument&) method.\n"
               << "Inputs number element is nullptr.\n";
 
        throw logic_error(buffer.str());
    }
 
    Index new_inputs_number;
 
    if(inputs_number_element->GetText())
    {
        new_inputs_number = static_cast<Index>(stoi(inputs_number_element->GetText()));
    }
 
    // Neurons number
 
    const tinyxml2::XMLElement* neurons_number_element = probabilistic_layer_element->FirstChildElement("NeuronsNumber");
 
    if(!inputs_number_element)
    {
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void from_XML(const tinyxml2::XMLDocument&) method.\n"
               << "Neurons number element is nullptr.\n";
 
        throw logic_error(buffer.str());
    }
 
    Index new_neurons_number;
 
    if(neurons_number_element->GetText())
    {
        new_neurons_number = static_cast<Index>(stoi(neurons_number_element->GetText()));
    }
 
    set(new_inputs_number, new_neurons_number);
 
    // Activation function
 
    const tinyxml2::XMLElement* activation_function_element = probabilistic_layer_element->FirstChildElement("ActivationFunction");
 
    if(!activation_function_element)
    {
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void from_XML(const tinyxml2::XMLDocument&) method.\n"
               << "Activation function element is nullptr.\n";
 
        throw logic_error(buffer.str());
    }
 
    if(activation_function_element->GetText())
    {
        set_activation_function(activation_function_element->GetText());
    }
 
    // Parameters
 
    const tinyxml2::XMLElement* parameters_element = probabilistic_layer_element->FirstChildElement("Parameters");
 
    if(!parameters_element)
    {
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void from_XML(const tinyxml2::XMLDocument&) method.\n"
               << "Parameters element is nullptr.\n";
 
        throw logic_error(buffer.str());
    }
 
    if(parameters_element->GetText())
    {
        const string parameters_string = parameters_element->GetText();
 
        set_parameters(to_type_vector(parameters_string, ' '));
    }
 
    // Decision threshold
 
    const tinyxml2::XMLElement* decision_threshold_element = probabilistic_layer_element->FirstChildElement("DecisionThreshold");
 
    if(!decision_threshold_element)
    {
        buffer << "OpenNN Exception: ProbabilisticLayer class.\n"
               << "void from_XML(const tinyxml2::XMLDocument&) method.\n"
               << "Decision threshold element is nullptr.\n";
 
        throw logic_error(buffer.str());
    }
 
    if(decision_threshold_element->GetText())
    {
        set_decision_threshold(static_cast<type>(atof(decision_threshold_element->GetText())));
    }
 
    // Display
 
    const tinyxml2::XMLElement* display_element = probabilistic_layer_element->FirstChildElement("Display");
 
    if(display_element)
    {
        const string new_display_string = display_element->GetText();
 
        try
        {
            set_display(new_display_string != "0");
        }
        catch(const logic_error& e)
        {
            cerr << e.what() << endl;
        }
    }
}
 
 
 
string ProbabilisticLayer::write_binary_expression(const Tensor<string, 1>& inputs_names, const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    buffer.str("");
 
    for(Index j = 0; j < outputs_names.size(); j++)
    {
        buffer << outputs_names(j) << " = binary(" << inputs_names(j) << ");\n";
    }
    return buffer.str();
}
 
 
 
string ProbabilisticLayer::write_logistic_expression(const Tensor<string, 1>& inputs_names,
                                                     const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    for(Index j = 0; j < outputs_names.size(); j++)
    {
        buffer << outputs_names(j) << " = logistic(" << inputs_names(j) << ");\n";
    }
    return buffer.str();
}
 
 
 
string ProbabilisticLayer::write_competitive_expression(const Tensor<string, 1>& inputs_names, const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    for(Index j = 0; j < outputs_names.size(); j++)
    {
        buffer << outputs_names(j) << " = competitive(" << inputs_names(j) << ");\n";
    }
    return buffer.str();
}
 
 
 
string ProbabilisticLayer::write_softmax_expression(const Tensor<string, 1>& inputs_names, const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    for(Index j = 0; j < outputs_names.size(); j++)
    {
        buffer << outputs_names(j) << " = softmax(" << inputs_names(j) << ");\n";
    }
 
    return buffer.str();
}
 
 
 
string ProbabilisticLayer::write_no_probabilistic_expression(const Tensor<string, 1>& inputs_names,
                                                             const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    for(Index j = 0; j < outputs_names.size(); j++)
    {
        buffer << outputs_names(j) << " = (" << inputs_names(j) << ");\n";
    }
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_combinations_c() const
{
    ostringstream buffer;
 
    const Index inputs_number = get_inputs_number();
    const Index neurons_number = get_neurons_number();
 
    buffer << "\tvector<float> combinations(" << neurons_number << ");\n" << endl;
 
    for(Index i = 0; i < neurons_number; i++)
    {
        buffer << "\tcombinations[" << i << "] = " << biases(i);
 
        for(Index j = 0; j < inputs_number; j++)
        {
            buffer << " +" << synaptic_weights(j, i) << "*inputs[" << j << "]";
        }
 
        buffer << ";" << endl;
    }
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_activations_c() const
{
    ostringstream buffer;
 
    const Index neurons_number = get_neurons_number();
 
    buffer << "\n\tvector<float> activations(" << neurons_number << ");\n" << endl;
 
    for(Index i = 0; i < neurons_number; i++)
    {
        switch(activation_function)
        {
        case ActivationFunction::Binary:
            buffer << "\tactivations[" << i << "] = combinations[" << i << "] < 0.5 ? 0.0 : 1.0;\n";
            break;
 
        case ActivationFunction::Logistic:
            buffer << "\tactivations[" << i << "] = 1.0/(1.0 + exp(-combinations[" << i << "]));\n";
            break;
 
        case ActivationFunction::Competitive:
 
            buffer << "";
 
            break;
 
        case ActivationFunction::Softmax:
 
            if(i == 0)
            {
                buffer << "\tfloat sum = 0;\n" << endl;
 
                buffer << "\tsum = ";
 
                for(Index i = 0; i < neurons_number; i++)
                {
                    buffer << "exp(combinations[" << i << "])";
 
                    if(i != neurons_number-1) buffer << " + ";
                }
 
                buffer << ";\n" << endl;
 
                for(Index i = 0; i < neurons_number; i++)
                {
                    buffer << "\tactivations[" << i << "] = exp(combinations[" << i << "])/sum;\n";
                }
            }
            break;
        }
    }
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_combinations_python() const
{
    ostringstream buffer;
 
    const Index inputs_number = get_inputs_number();
    const Index neurons_number = get_neurons_number();
 
    buffer << "\t\tcombinations = [None] * "<<neurons_number<<"\n" << endl;
 
    for(Index i = 0; i < neurons_number; i++)
    {
        buffer << "\t\tcombinations[" << i << "] = " << biases(i);
 
        for(Index j = 0; j < inputs_number; j++)
        {
            buffer << " +" << synaptic_weights(j, i) << "*inputs[" << j << "]";
        }
 
        buffer << " " << endl;
    }
 
    buffer << "\t\t" << endl;
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_activations_python() const
{
    ostringstream buffer;
 
    const Index neurons_number = get_neurons_number();
 
    buffer << "\t\tactivations = [None] * "<<neurons_number<<"\n" << endl;
 
    for(Index i = 0; i < neurons_number; i++)
    {
        switch(activation_function)
        {
        case ActivationFunction::Binary:
            buffer << "\t\tactivations[" << i << "] = 0.0 if combinations[" << i << "] < 0.5 else 1.0\n";
            break;
 
        case ActivationFunction::Logistic:
            buffer << "\t\tactivations[" << i << "] = 1.0/(1.0 + np.exp(-combinations[" << i << "]));\n";
            break;
 
        case ActivationFunction::Competitive:
 
            if(i == 0)
            {
                buffer << "\t\tfor i, value in enumerate(combinations):"<<endl;
 
                buffer <<"\t\t\tif(max(combinations) == value):"<<endl;
 
                buffer <<"\t\t\t\tactivations[i] = 1"<<endl;
 
                buffer <<"\t\t\telse:"<<endl;
 
                buffer <<"\t\t\t\tactivations[i] = 0"<<endl;
            }
 
            break;
 
        case ActivationFunction::Softmax:
 
            if(i == 0)
            {
                buffer << "\t\tsum_ = 0;\n" << endl;
 
                buffer << "\t\tsum_ = ";
 
                for(Index i = 0; i < neurons_number; i++)
                {
                    buffer << "\tnp.exp(combinations[" << i << "])";
 
                    if(i != neurons_number-1) buffer << " + ";
                }
 
                buffer << ";\n" << endl;
 
                for(Index i = 0; i < neurons_number; i++)
                {
                    buffer << "\t\tactivations[" << i << "] = np.exp(combinations[" << i << "])/sum_;\n";
                }
 
            }
            break;
        }
    }
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_combinations(const Tensor<string, 1>& inputs_names) const
{
    ostringstream buffer;
 
    const Index inputs_number = get_inputs_number();
    const Index neurons_number = get_neurons_number();
 
    for(Index i = 0; i < neurons_number; i++)
    {
        buffer << "probabilistic_layer_combinations_" << to_string(i) << " = " << biases(i);
 
        for(Index j = 0; j < inputs_number; j++)
        {
            buffer << " +" << synaptic_weights(j, i) << "*" << inputs_names(j) << "";
        }
 
        buffer << " " << endl;
    }
 
    buffer << "\t" << endl;
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_activations(const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    const Index neurons_number = get_neurons_number();
 
    for(Index i = 0; i < neurons_number; i++)
    {
        switch(activation_function)
        {
        case ActivationFunction::Binary:
        {
            buffer << "\tif" << "probabilistic_layer_combinations_" << to_string(i) << " < 0.5, " << outputs_names(i) << "= 0.0. Else " << outputs_names(i) << " = 1.0\n";
        }
            break;
 
        case ActivationFunction::Logistic:
        {
            buffer <<  outputs_names(i) << " = 1.0/(1.0 + exp(-" <<  "probabilistic_layer_combinations_" << to_string(i) << ");\n";
        }
            break;
 
        case ActivationFunction::Competitive:
            if(i == 0)
            {
                buffer << "\tfor each probabilistic_layer_combinations_i:"<<endl;
 
                buffer <<"\t\tif probabilistic_layer_combinations_i is equal to max(probabilistic_layer_combinations_i):"<<endl;
 
                buffer <<"\t\t\tactivations[i] = 1"<<endl;
 
                buffer <<"\t\telse:"<<endl;
 
                buffer <<"\t\t\tactivations[i] = 0"<<endl;
            }
 
            break;
 
        case ActivationFunction::Softmax:
 
            if(i == 0)
            {
                buffer << "sum = ";
 
                for(Index i = 0; i < neurons_number; i++)
                {
                    buffer << "exp(probabilistic_layer_combinations_" << to_string(i) << ")";
 
                    if(i != neurons_number-1) buffer << " + ";
                }
 
                buffer << ";\n" << endl;
 
                for(Index i = 0; i < neurons_number; i++)
                {
                    buffer << outputs_names(i) << " = exp(probabilistic_layer_combinations_" << to_string(i) <<")/sum;\n";
                }
 
            }
            break;
        }
    }
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_expression_c() const
{
    ostringstream buffer;
 
    buffer << "vector<float> " << layer_name << "(const vector<float>& inputs)\n{" << endl;
 
    buffer << write_combinations_c();
 
    buffer << write_activations_c();
 
    buffer << "\n\treturn activations;\n}" << endl;
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_expression_python() const
{
    ostringstream buffer;
 
    buffer << "\tdef " << layer_name << "(self, inputs):\n" << endl;
 
    buffer << write_combinations_python();
 
    buffer << write_activations_python();
 
    buffer << "\n\t\treturn activations;\n" << endl;
 
    return buffer.str();
}
 
 
string ProbabilisticLayer::write_expression(const Tensor<string, 1>& inputs_names,
                                            const Tensor<string, 1>& outputs_names) const
{
    ostringstream buffer;
 
    buffer << write_combinations(inputs_names);
 
    buffer << write_activations(outputs_names);
 
    return buffer.str();
}
 
 
 
}
 
// OpenNN: Open Neural Networks Library.
// Copyright(C) 2005-2021 Artificial Intelligence Techniques, SL.
//
// This library is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 2.1 of the License, or any later version.
//
// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
// Lesser General Public License for more details.
 
// You should have received a copy of the GNU Lesser General Public
// License along with this library; if not, write to the Free Software
// Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA