ACSFNeuralNetwork.cpp

//
// Created by martin on 19.08.19.
//

#include <string>

#include "../exceptions.h"
#include "../settings.h"
#include "ACSFNeuralNetwork.h"
#include "../ErrorFunction/ErrorFunctions.h"

void lib4neuro::ACSFNeuralNetwork::load_structure_from_file(const char* filepath){
    this->init_from_file( filepath );
}

lib4neuro::ACSFNeuralNetwork::ACSFNeuralNetwork(std::unordered_map<ELEMENT_SYMBOL, Element*>& elements,
                                                std::vector<ELEMENT_SYMBOL>& elements_list,
                                                bool with_charge,
                                                std::unordered_map<ELEMENT_SYMBOL, std::vector<unsigned int>> n_hidden_neurons,
                                                std::unordered_map<ELEMENT_SYMBOL, std::vector<NEURON_TYPE>> type_hidden_neurons) {

    /* Check parameters */
    for(auto symbol : elements_list) {
        if(n_hidden_neurons[symbol].size() != type_hidden_neurons[symbol].size()) {
            THROW_RUNTIME_ERROR("Number of hidden layers for " + elements[symbol]->getElementSymbol() + " ("
                                + std::to_string(n_hidden_neurons[symbol].size())
                                + ") doesn't correspond with a number of hidden neuron types ("
                                + std::to_string(type_hidden_neurons[symbol].size()) + ")!");
        }
    }

    /* Save info about elements */
    this->elements = &elements;
    this->elements_list = &elements_list;

    /* Construct the neural network */
    std::vector<size_t> inputs;

    std::unordered_map<ELEMENT_SYMBOL, size_t> subnet_neuron_shifts;
    std::unordered_map<ELEMENT_SYMBOL, size_t> subnet_connection_shifts;
    std::unordered_map<ELEMENT_SYMBOL, bool> subnet_constructed;

	size_t last_neuron_bias_idx = 0;
	size_t last_connection_weight_idx = 0;

	std::shared_ptr<Neuron> output_neuron = std::make_shared<NeuronLinear>();
	size_t last_neuron_idx = this->add_neuron(output_neuron, BIAS_TYPE::NO_BIAS);
    std::vector<size_t> outputs = {last_neuron_idx};

    for(size_t i = 0; i < elements_list.size(); i++) {
        std::vector<size_t> previous_layer;
        std::vector<size_t> new_layer;

		size_t first_input_neuron_index = last_neuron_idx + 1;

        /* Create input neurons for sub-net */
        std::shared_ptr<NeuronLinear> inp_n;
        for(size_t j = 0; j < elements[elements_list.at(i)]->getSymmetryFunctions()->size(); j++) {
            inp_n = std::make_shared<NeuronLinear>();
            last_neuron_idx = this->add_neuron(inp_n, BIAS_TYPE::NO_BIAS);
            previous_layer.emplace_back(last_neuron_idx);
            inputs.emplace_back(last_neuron_idx);
        }

        /* Add an additional input neuron for charge, if provided */
        if(with_charge) {
            inp_n = std::make_shared<NeuronLinear>();
            last_neuron_idx = this->add_neuron(inp_n, BIAS_TYPE::NO_BIAS);
            previous_layer.emplace_back(last_neuron_idx);
            inputs.emplace_back(last_neuron_idx);
        }

        /* Create subnet for the current element */
        bool new_subnet = false;
        if(subnet_constructed.find(elements_list.at(i)) == subnet_constructed.end()) {
            new_subnet = true;
			subnet_constructed[elements_list.at(i)] = true;
			subnet_neuron_shifts[elements_list.at(i)] = last_neuron_bias_idx;
			subnet_connection_shifts[elements_list.at(i)] = last_connection_weight_idx;
        }

		// std::cout << "Particle " << i << ", input neuron indices: " << first_input_neuron_index << " - " << last_neuron_idx << std::endl;

        /* Create hidden layers in sub-net */
        std::vector<unsigned int> n_neurons = n_hidden_neurons[ elements_list.at(i) ];
        std::vector<NEURON_TYPE> types = type_hidden_neurons[ elements_list.at(i) ];

        size_t local_neuron_idx = subnet_neuron_shifts[elements_list.at(i)];
        size_t local_connection_idx = subnet_connection_shifts[elements_list.at(i)];

        for(size_t j = 0; j < n_neurons.size(); j++) { /* Iterate over hidden layers */
            /* Create hidden neurons */
            for(size_t k = 0; k < n_neurons.at(j); k++) {
                std::shared_ptr<Neuron> hid_n;
                switch(types.at(j)) {
                    case NEURON_TYPE::LOGISTIC: {
                        hid_n = std::make_shared<NeuronLogistic>();
                        break;
                    }
                    case NEURON_TYPE::BINARY: {
                        hid_n = std::make_shared<NeuronBinary>();
                        break;
                    }
                    case NEURON_TYPE::CONSTANT: {
                        hid_n = std::make_shared<NeuronConstant>();
                        break;
                    }
                    case NEURON_TYPE::LINEAR: {
                        hid_n = std::make_shared<NeuronLinear>();
                        break;
                    }
                }

                if(new_subnet) {
                    last_neuron_idx = this->add_neuron(hid_n,
                                                  BIAS_TYPE::NEXT_BIAS);
					// std::cout << "  new subnet, neuron index: " << last_neuron_idx << ", neuron bias: " << last_neuron_bias_idx << std::endl;
					last_neuron_bias_idx++;
                } else {
                    last_neuron_idx = this->add_neuron(hid_n,
                                                  BIAS_TYPE::EXISTING_BIAS, local_neuron_idx);
					// std::cout << "  old subnet, neuron index: " << last_neuron_idx << ", neuron bias: " << local_neuron_idx << std::endl;
                }
                local_neuron_idx++;
                new_layer.emplace_back(last_neuron_idx);

                /* Connect hidden neuron to the previous layer */
                for(auto prev_n : previous_layer) {
                    if(new_subnet) {
                        this->add_connection_simple(prev_n,
                                                    last_neuron_idx,
                                                    SIMPLE_CONNECTION_TYPE::NEXT_WEIGHT);
						// std::cout << "    new subnet, connection weight bias: " << last_connection_weight_idx << std::endl;
						last_connection_weight_idx++;
                    } else {
                        this->add_connection_simple(prev_n,
                                                    last_neuron_idx,
                                                    SIMPLE_CONNECTION_TYPE::EXISTING_WEIGHT, local_connection_idx);
						// std::cout << "    old subnet, connection weight bias: " << local_connection_idx << std::endl;
                    }
                    local_connection_idx++;
                }
            }
            previous_layer = new_layer;
            new_layer.clear();
        }

        /* Create output neurons for sub-net */
        for(auto prev_n : previous_layer) {
            this->add_connection_constant(prev_n, outputs[ 0 ], 1.0);
        }
    }

    /* Specify network inputs and outputs */
//    COUT_INFO("input len: " << inputs.size() );
    this->specify_input_neurons(inputs);
    this->specify_output_neurons(outputs);
}

std::unordered_map<lib4neuro::ELEMENT_SYMBOL, lib4neuro::Element*>* lib4neuro::ACSFNeuralNetwork::get_elements() {
    return this->elements;
}

std::vector<lib4neuro::ELEMENT_SYMBOL>* lib4neuro::ACSFNeuralNetwork::get_elements_list() {
    return this->elements_list;
}