nodes_value_containers.hpp

// Copyright 2021 Thomas A. R. Purcell
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

/** @file feature_creation/node/value_storage/node_value_containers.hpp
 *  @brief Creates a central storage area for the feature's input data and the descriptor matrix
 *
 *  @author Thomas A. R. Purcell (tpurcell90)
 *  @bug No known bugs.
 */

#ifndef NODE_VALUE_ARR
#define NODE_VALUE_ARR

#include <algorithm>
#include <cmath>
#include <memory>
#include <numeric>
#include <vector>
#include <iostream>

#include <omp.h>

#include"utils/enum.hpp"

#ifdef PY_BINDINGS
#include "python/py_binding_cpp_def/conversion_utils.hpp"
namespace np = boost::python::numpy;
namespace py = boost::python;
#endif

namespace node_value_arrs
{
    extern std::vector<double> VALUES_ARR; //!< The central storage location for the values of the training data for each feature with a rung <= N_RUNGS_STORED
    extern std::vector<double> TEST_VALUES_ARR; //!< The central storage location for the values of the test data for each feature with a rung <= N_RUNGS_STORED

    extern std::vector<double> TEMP_STORAGE_ARR; //!< The vector used to temporarily store the values of each feature with a rung > N_RUNGS_STORED (These are calculated on the fly when the values are required)
    extern std::vector<double> TEMP_STORAGE_TEST_ARR; //!< The vector used to temporarily store the values of each feature with a rung > N_RUNGS_STORED (These are calculated on the fly when the values are required)
    extern std::vector<int> TEMP_STORAGE_REG; //!< Register that maps the slots in TEMP_STORAGE_ARR to the index of the feature whose data is currently stored there (reset by setting all elements of this vector to -1)
    extern std::vector<int> TEMP_STORAGE_TEST_REG; //!< Register that maps the slots in TEMP_STORAGE_TEST_ARR to the index of the feature whose data is currently stored there (reset by setting all elements of this vector to -1)

    extern std::vector<double> PARAM_STORAGE_ARR; //!< The vector used to temporarily store the values of each feature that has free-parameters (These are calculated on the fly when the values are required)
    extern std::vector<double> PARAM_STORAGE_TEST_ARR; //!< The vector used to temporarily store the values of each feature that has free-parameters (These are calculated on the fly when the values are required)

    extern std::vector<double> D_MATRIX; //!< The descriptor matrix (Central storage for the selected feature space)

    extern std::vector<int> TASK_SZ_TRAIN; //!< Number of training samples per task
    extern std::vector<int> TASK_START_TRAIN; //!< The starting point for each task in the training data
    extern std::vector<int> TASK_SZ_TEST; //!< Number of test sample per task

    extern int N_SELECTED; //!< Number of selected features

    extern int N_SAMPLES; //!< Number of training samples for each feature (Sum of all elements in TASK_SZ_TRAIN)
    extern int N_SAMPLES_TEST; //!< Number of test samples for each feature (Sum of all elements in TASK_SZ_TEST)

    extern int N_PRIMARY_FEATURES; //!< Number of primary features
    extern int N_STORE_FEATURES; //!< Number of features with stored values
    extern int N_RUNGS_STORED; //!< Maximum rung for permanently storing a features value
    extern int MAX_RUNG; //!< The maximum rung for all features

    extern int MAX_N_THREADS; //!< Get the maximum number of threads possible for a calculation
    extern int N_OP_SLOTS; //!< The number of possible nodes of the binary expression tree that maybe calculated on the fly
    extern int N_PARAM_OP_SLOTS; //!< The number of possible non-leaf nodes of the binary expression tree
    /**
     * @brief Initialize all central storage vectors/descriptive variables without changing MAX_RUNG
     *
     * @param n_samples The number of training samples for each feature (Sum of all elements in TASK_SZ_TRAIN)
     * @param n_samples_test The number of test samples for each feature (Sum of all elements in TASK_SZ_TEST)
     * @param n_primary_feat The number of primary features
     */
    void initialize_values_arr(
        const int n_samples,
        const int n_samples_test,
        const int n_primary_feat
    );

    /**
     * @brief Initialize all central storage vectors/descriptive variables
     *
     * @param task_sz_train The number of training samples per task
     * @param task_sz_test The number of test sample per task
     * @param n_primary_feat The number of primary features
     * @param max_rung The maximum rung for all features
     * @param use_params If True set up parameterized feature storage as well
     */
    void initialize_values_arr(
        const std::vector<int> task_sz_train,
        const std::vector<int> task_sz_test,
        const int n_primary_feat,
        const int max_rung,
        const bool use_params
    );

    /**
     * @brief Resize the central storage array given a new number of features and the current rung of the features
     *
     * @param n_rung The current rung of the generated features
     * @param n_feat The new number of features to store
     */
    void resize_values_arr(const int n_dims, const int n_feat);

    // DocString: node_vals_init_param
    /**
     * @brief Initialize the parameter storage array
     */

    void initialize_param_storage();

    // DocString: node_vals_init_d_mat
    /**
     * @brief Initialize the descriptor matrix
     *
     */
    void initialize_d_matrix_arr();

    /**
     * @brief Resize the descriptor matrix for the new number of selected features
     *
     * @param n_select Number of features to select
     */
    void resize_d_matrix_arr(const int n_select);

    /**
     * @brief Reset the global TASK_SZ_TRAIN vector
     *
     * @param task_sz_train the new task_sz train
     */
    void set_task_sz_train(const std::vector<int> task_sz_train);

    /**
     * @brief Reset the global TASK_SZ_TEST vector
     *
     * @param task_sz_train the new test_sz train
     */
    void set_task_sz_test(const std::vector<int> task_sz_test);

    /**
     * @brief Set max_rung and initialize the temporary storage arrays
     *
     * @param max_rung The maximum rung for the calculation
     */
    void set_max_rung(const int max_rung, bool use_params=false);

    /**
     * @brief Get the operator slot associated with a given rung/offset
     *
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     *
     * @return The operator slot to use
     */
    inline int get_op_slot(const int rung, const int offset, const bool for_comp)
    {
        return std::abs(N_OP_SLOTS / (1 + !for_comp) - static_cast<int>(std::pow(2, MAX_RUNG - rung)) - offset);
    }

    /**
     * @brief Get the parameter operator slot associated with a given rung/offset
     *
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     *
     * @return The operator slot to use
     */
    inline int get_param_op_slot(const int rung, const int offset, const bool for_comp)
    {
        return std::abs(N_PARAM_OP_SLOTS / (1 + !for_comp) - static_cast<int>(std::pow(2, MAX_RUNG - rung)) - offset);
    }

    /**
     * @brief Get a reference slot/feature register of the training data
     *
     * @param ind The Node's arr_ind
     * @param op_slot(int) Offset integer for TEMP_STORE_ARRAY
     *
     * @return The register element for a given feature index and op_slot
     */
    inline int& temp_storage_reg(const unsigned long int ind, const int op_slot=0)
    {
        return TEMP_STORAGE_REG[
            (ind % N_PRIMARY_FEATURES) + (op_slot % N_OP_SLOTS) * N_PRIMARY_FEATURES + omp_get_thread_num() * (N_PRIMARY_FEATURES * N_OP_SLOTS + 1)
        ];
    }

    /**
     * @brief Get a reference slot/feature register of the test data
     *
     * @param ind The Node's arr_ind
     * @param op_slot(int) Offset integer for TEMP_STORE_TEST_ARRAY
     *
     * @return The register element for a given feature index and op_slot
     */
    inline int& temp_storage_test_reg(const unsigned long int ind, const int op_slot=0)
    {
        return TEMP_STORAGE_TEST_REG[
            (ind % N_PRIMARY_FEATURES) + (op_slot % N_OP_SLOTS) * N_PRIMARY_FEATURES + omp_get_thread_num() * (N_PRIMARY_FEATURES * N_OP_SLOTS + 1)
        ];
    }

    /**
     * @brief Get a reference slot/feature register of the training data
     *
     * @param ind The Node's arr_ind
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     *
     * @return The register element for a given feature index and offset
     */
    inline int& temp_storage_reg(const unsigned long int ind, const int rung, const int offset, const bool for_comp)
    {
        return TEMP_STORAGE_REG[
            (ind % N_PRIMARY_FEATURES) +
            (get_op_slot(rung, offset, for_comp) % N_OP_SLOTS) * N_PRIMARY_FEATURES +
            omp_get_thread_num() * (N_PRIMARY_FEATURES * N_OP_SLOTS + 1)
        ];
    }

    /**
     * @brief Get a reference slot/feature register of the test data
     *
     * @param ind The Node's arr_ind
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     *
     * @return The register element for a given feature index and offset
     */
    inline int& temp_storage_test_reg(const unsigned long int ind, const int rung, const int offset, const bool for_comp)
    {
        return TEMP_STORAGE_TEST_REG[
            (ind % N_PRIMARY_FEATURES) +
            (get_op_slot(rung, offset, for_comp) % N_OP_SLOTS) * N_PRIMARY_FEATURES +
            omp_get_thread_num() * (N_PRIMARY_FEATURES * N_OP_SLOTS + 1)
        ];
    }

    /**
     * @brief Access element of the permanent training data storage array
     *
     * @param feature_ind The _feat_ind of Node to get the training data of
     *
     * @return pointer to the Node's training data
     */
    inline double* access_value_arr(const unsigned long int feature_ind){return &VALUES_ARR[feature_ind*N_SAMPLES];}

    /**
     * @brief Access element of the permanent test data storage array
     *
     * @param feature_ind The _feat_ind of Node to get the test data of
     *
     * @return pointer to the Node's test data
     */
    inline double* access_test_value_arr(const unsigned long int feature_ind){return &TEST_VALUES_ARR[feature_ind*N_SAMPLES_TEST];}

    /**
     * @brief Access element of temporary storage array for the training data
     *
     * @param slot The slot of the temporary storage array
     *
     * @return pointer to the data stored in the specified slot
     */
    inline double* access_temp_storage(const unsigned long int slot){return &TEMP_STORAGE_ARR[slot*N_SAMPLES];}

    /**
     * @brief Access element of temporary storage array for the test data
     *
     * @param slot The slot of the temporary storage array
     *
     * @return pointer to the data stored in the specified slot
     */
    inline double* access_temp_storage_test(const unsigned long int slot){return &TEMP_STORAGE_TEST_ARR[slot*N_SAMPLES_TEST];}

    /**
     * @brief Access the param storage array
     *
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     * @return pointer to the correct element of the PARAM_STORAGE_ARR
     */
    inline double* access_param_storage(const int rung=0, const int offset=0, const bool for_comp=false)
    {
        return &PARAM_STORAGE_ARR[
            N_SAMPLES *
            (
                (get_param_op_slot(rung, offset, for_comp) % N_PARAM_OP_SLOTS) +
                omp_get_thread_num() * (N_PARAM_OP_SLOTS + 1)
            )
        ];
    }

    /**
     * @brief Access the param storage array for the test set
     *
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     * @return pointer to the correct element of the PARAM_STORAGE_ARR
     */
    inline double* access_param_storage_test(const int rung=0, const int offset=0, const bool for_comp=false)
    {
        return &PARAM_STORAGE_TEST_ARR[
            N_SAMPLES_TEST *
            (
                (get_param_op_slot(rung, offset, for_comp) % N_PARAM_OP_SLOTS) +
                omp_get_thread_num() * (N_PARAM_OP_SLOTS + 1)
            )
        ];
    }

    /**
     * @brief Get a Node's value_ptr
     *
     * @param arr_ind Nodes _arr_ind
     * @param feat_ind Nodes _feat_ind
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     *
     * @return The value pointer
     */
    double* get_value_ptr(
        const unsigned long int arr_ind,
        const unsigned long int feat_ind,
        const int rung=0,
        const int offset=0,
        const bool for_comp=false
    );

    /**
     * @brief Get a Node's test_value_ptr
     *
     * @param arr_ind Nodes _arr_ind
     * @param feat_ind Nodes _feat_ind
     * @param rung Rung of the feature
     * @param offset Offset used to prevent overwrites (determines where in the binary expression tree this operator is)
     * @param for_comp If true get a slot dedicated to comparing features
     *
     * @return The value pointer
     */
    double* get_test_value_ptr(
        const unsigned long int arr_ind,
        const unsigned long int feat_ind,
        const int rung=0,
        const int offset=0,
        const bool for_comp=false
    );

    /**
     * @brief Get the pointer to a particular selected Node from sis
     *
     * @param ind Index of the data in the descriptor matrix
     * @return The pointer to the descriptor matrix's data
     */
    inline double* get_d_matrix_ptr(const int ind){return &D_MATRIX[ind * N_SAMPLES];}

    /**
     * @brief Get the pointer to a particular selected Node from sis
     *
     * @param ind Index of the data in the descriptor matrix
     * @param taskind The index for the given task
     * @return The pointer to the descriptor matrix's data
     */
    inline double* get_d_matrix_ptr(const int ind, const int taskind){return &D_MATRIX[ind * N_SAMPLES + TASK_START_TRAIN[taskind]];}

    /**
     * @brief Flush the temporary storage register (training data)
     * @details Reset all slots in the register to -1
     */
    inline void clear_temp_reg(){std::fill_n(TEMP_STORAGE_REG.begin(), TEMP_STORAGE_REG.size(), -1);}

    /**
     * @brief Flush the temporary storage register (training data)
     * @details Reset all slots in the register to -1
     */
    inline void clear_temp_reg_thread()
    {
        std::fill_n(TEMP_STORAGE_REG.begin() + (N_PRIMARY_FEATURES * N_OP_SLOTS + 1) * omp_get_thread_num(), N_PRIMARY_FEATURES * N_OP_SLOTS + 1, -1);
    }

    /**
     * @brief Flush the temporary storage register (test data)
     * @details Reset all slots in the register to -1
     */
    inline void clear_temp_test_reg(){std::fill_n(TEMP_STORAGE_TEST_REG.begin(), TEMP_STORAGE_TEST_REG.size(), -1);}

#ifdef PY_BINDINGS

    // DocString: node_vals_ts_list_no_params
    /**
     * @brief Initialize the node value arrays
     * @details Using the size of the initial feature space constructor the storage arrays
     *
     * @param task_sz_train (list) The number of training samples per task
     * @param task_sz_test (list) The number of test sample per task
     * @param n_primary_feat (int) The number of primary features
     * @param max_rung (int) The maximum rung for all features
     */
    inline void initialize_values_arr(
        py::list task_sz_train,
        py::list task_sz_test,
        int n_primary_feat,
        int max_rung
    )
    {
        initialize_values_arr(
            python_conv_utils::from_list<int>(task_sz_train),
            python_conv_utils::from_list<int>(task_sz_test),
            n_primary_feat,
            max_rung,
            false
        );
    }

    // DocString: node_vals_ts_arr_no_params
    /**
     * @brief Initialize the node value arrays
     * @details Using the size of the initial feature space constructor the storage arrays
     *
     * @param task_sz_train (np.ndarray) The number of training samples per task
     * @param task_sz_test (np.ndarray) The number of test sample per task
     * @param n_primary_feat (int) The number of primary features
     * @param max_rung (int) The maximum rung for all features
     */
    inline void initialize_values_arr(
        np::ndarray task_sz_train,
        np::ndarray task_sz_test,
        int n_primary_feat,
        int max_rung
    )
    {
        initialize_values_arr(
            python_conv_utils::from_ndarray<int>(task_sz_train),
            python_conv_utils::from_ndarray<int>(task_sz_test),
            n_primary_feat,
            max_rung,
            false
        );
    }

    // DocString: node_vals_ts_list
    /**
     * @brief Initialize the node value arrays
     * @details Using the size of the initial feature space constructor the storage arrays
     *
     * @param task_sz_train (list): The number of training samples per task
     * @param task_sz_test (list): The number of test sample per task
     * @param n_primary_feat (int): The number of primary features
     * @param max_rung (int): The maximum rung for all features
     * @param use_params (bool): If true also initialize parameterized storage
     */
    inline void initialize_values_arr(
        py::list task_sz_train,
        py::list task_sz_test,
        int n_primary_feat,
        int max_rung,
        bool use_params
    )
    {
        initialize_values_arr(
            python_conv_utils::from_list<int>(task_sz_train),
            python_conv_utils::from_list<int>(task_sz_test),
            n_primary_feat,
            max_rung,
            use_params
        );
    }

    // DocString: node_vals_ts_arr
    /**
     * @brief Initialize the node value arrays
     * @details Using the size of the initial feature space constructor the storage arrays
     *
     * @param task_sz_train (np.ndarray): The number of training samples per task
     * @param task_sz_test (np.ndarray): The number of test sample per task
     * @param n_primary_feat (int): The number of primary features
     * @param max_rung (int): The maximum rung for all features
     * @param use_params (bool): If true also initialize parameterized storage
     */
    inline void initialize_values_arr(
        np::ndarray task_sz_train,
        np::ndarray task_sz_test,
        int n_primary_feat,
        int max_rung,
        bool use_params
    )
    {
        initialize_values_arr(
            python_conv_utils::from_ndarray<int>(task_sz_train),
            python_conv_utils::from_ndarray<int>(task_sz_test),
            n_primary_feat,
            max_rung,
            use_params
        );
    }
#endif
}

#endif