// 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/feature_space/FeatureSpace.hpp
 *  @brief Defines the class for creating/operating on a feature space in SISSO
 *
 *  @author Thomas A. R. Purcell (tpurcell90)
 *  @bug No known bugs.
 */

#ifndef FEATURE_SPACE
#define FEATURE_SPACE

#include <boost/filesystem.hpp>

#include <utility>

#include "feature_creation/node/utils.hpp"
#include "inputs/InputParser.hpp"

#include "mpi_interface/MPI_Interface.hpp"
#include "mpi_interface/MPI_Ops.hpp"
#include "mpi_interface/serialize_tuple.h"

#include "utils/project.hpp"
#include "mpi_interface/MPI_Interface.hpp"

#ifdef PY_BINDINGS
    namespace np = boost::python::numpy;
    namespace py = boost::python;
#endif

// DocString: cls_feat_space
/**
 * @brief Feature Space for SISSO calculations. It stores and performs all actions on the feature space for SISSO.
 *
 */
class FeatureSpace
{
    std::vector<node_ptr> _phi_selected; //!< A vector containing all of the selected features
    std::vector<node_ptr> _phi; //!< A vector containing all features generated (Not including those created on the Fly during SIS)
    std::vector<node_ptr> _phi_0; //!< A vector containing all of the Primary features

    #ifdef PARAMETERIZE
    std::vector<node_ptr> _phi_reparam; //!< A vector containing the features created when reparameterizating using the residuals
    std::vector<int> _end_no_params; //!< A vector containing the indexes of each rung where parameterized nodes start
    std::vector<int> _start_rung_reparam; //!< A vector containing the indexes of each rung where parameterized nodes start

    std::vector<un_param_op_node_gen> _un_param_operators; //!< Vector containing all parameterized unary operators with free parameters
    std::vector<bin_param_op_node_gen> _com_bin_param_operators; //!< Vector containing all parameterized commutable binary operators with free parameters
    std::vector<bin_param_op_node_gen> _bin_param_operators; //!< Vector containing all parameterized binary operators with free parameters
    std::vector<std::string> _allowed_param_ops; //!< Vector containing all allowed operators strings for operators with free parameters
    #endif

    std::vector<std::string> _allowed_ops; //!< Vector containing all allowed operators strings
    std::vector<un_op_node_gen> _un_operators; //!< Vector containing all unary operators
    std::vector<bin_op_node_gen> _com_bin_operators; //!< Vector containing all commutable binary operators
    std::vector<bin_op_node_gen> _bin_operators; //!< Vector containing all binary operators

    std::vector<double> _prop_train; //!< The value of the property vector for each training sample
    std::vector<double> _scores; //!< The projection scores for each feature

    const std::vector<int> _task_sizes_train; //!< Number of training samples per task
    std::vector<int> _start_rung; //!< Vector containing the indexes where each rung starts in _phi
    const std::string _project_type; //!< The type of LossFunction to use when projecting the features onto a property
    const std::string _feature_space_file; //!< File to output the computer readable representation of the selected features to
    const std::string _feature_space_summary_file; //!< File to output the human readable representation of the selected features to
    const std::string _phi_out_file; //!< Filename of the file to output the feature set to

    std::function<bool(const double*, const int, const double, const std::vector<double>&, const double, const int, const int)> _is_valid; //!< Function used to determine of a feature is too correlated to previously selected features
    std::function<int(const double*, const int, const double, const std::vector<node_ptr>&, const std::vector<double>&, const double)> _is_valid_feat_list; //!< Function used to determine of a feature is too correlated to previously selected features within a given list

    std::shared_ptr<MPI_Interface> _mpi_comm; //!< the MPI communicator for the calculation

    const double _cross_cor_max; //!< Maximum cross-correlation used for selecting features
    const double _l_bound; //!< The lower bound for the maximum absolute value of the features
    const double _u_bound; //!< The upper bound for the maximum absolute value of the features

    int _n_rung_store; //!< The number of rungs to calculate and store the value of the features for all samples
    int _n_feat; //!< Total number of features in the feature space
    int _max_rung; //!< Maximum rung for the feature creation

    const int _n_sis_select; //!< Number of features to select during each SIS iteration
    const int _n_samp_train; //!< Number of samples in the training set
    const int _n_rung_generate; //!< Either 0 or 1, and is the number of rungs to generate on the fly during SIS

    #ifdef PARAMETERIZE
    int _max_param_depth; //!< The maximum depth in the binary expression tree to set non-linear optimization
    bool _reparam_residual; //!< If True then reparameterize features using the residuals of each model
    #endif
public:

    // DocString: feat_space_init
    /**
     * @brief Construct a FeatureSpace using an InputParser object
     *
     * @param inputs InputParser object used to build the FeatureSpace
     */
    FeatureSpace(InputParser inputs);

    /**
     * @brief FeatureSpace constructor that uses a file containing postfix feature expressions to describe all features in Phi, and a primary feature setn <python/feature_creation/FeatureSpace.cpp>)
     *
     * @param feature_file The file containing the postfix expressions of all features in the FeatureSpace
     * @param phi_0 The set of primary features
     * @param prop List containing the property vector (training data only)
     * @param task_sizes_train The number of samples in the training data per task
     * @param project_type The type of loss function/projection operator to use
     * @param n_sis_select The number of features to select during each SIS step
     * @param cross_corr_max The maximum allowed cross-correlation value between selected features
     */
    FeatureSpace(
        std::string feature_file,
        std::vector<node_ptr> phi_0,
        std::vector<double> prop,
        std::vector<int> task_sizes_train,
        std::string project_type="regression",
        int n_sis_select=1,
        double cross_corr_max=1.0,
        std::vector<int> excluded_inds = std::vector<int>()
    );

    /**
     * @brief Destructor
     */
    ~FeatureSpace();

    /**
     * @brief Populate the operator lists using _allowed_ops and _allowed_param_ops
     */
    void set_op_lists();

    /**
     * @brief Create SIS output files and write their headers
     */
    void initialize_fs_output_files() const;

    /**
     * @brief Populate _phi using _phi_0 and the allowed operators up to (_max_rung - _n_rung_generate)^th rung
     */
    void generate_feature_space(
        std::vector<node_ptr>& feat_set,
        std::vector<int>& start_rung,
        const std::vector<double>& prop,
        bool reparam = false
    );

    /**
     * @brief A vector containing all of the selected features
     */
    inline std::vector<node_ptr> phi_selected() const {return _phi_selected;};

    /**
     * @brief A vector containing all features generated (Not including those created on the Fly during SIS)
     */
    inline std::vector<node_ptr> phi() const {return _phi;};

    /**
     * @brief A vector containing all of the Primary features
     */
    inline std::vector<node_ptr> phi0() const {return _phi_0;};

    /**
     * @brief The projection scores for each feature in _phi
     */
    inline std::vector<double> scores() const {return _scores;}

    /**
     * @brief The MPI Communicator
     */
    inline std::shared_ptr<MPI_Interface> mpi_comm() const {return _mpi_comm;}

    /**
     * @brief Number of training samples per task
     */
    inline std::vector<int> task_sizes_train() const {return _task_sizes_train;}

    // DocString: feat_space_feature_space_file
    /**
     * @brief Filename of the file to output the computer readable representation of the selected features to
     */
    inline std::string feature_space_file() const {return _feature_space_file;}

    // DocString: feat_space_feature_space_file
    /**
     * @brief Filename of the file to output the human readable representation of the selected features to
     */
    inline std::string feature_space_summary_file() const {return _feature_space_summary_file;}

    // DocString: feat_space_l_bound
    /**
     * @brief The mlower bound for the maximum absolute value of the features
     */
    inline double l_bound() const {return _l_bound;}

    // DocString: feat_space_u_bound
    /**
     * @brief The upper bound for the maximum absolute value of the features
     */
    inline double u_bound() const {return _u_bound;}

    // DocString: feat_space_max_rung
    /**
     * @brief The maximum rung for the feature creation
     */
    inline int max_rung() const {return _max_rung;}

    // DocString: feat_space_n_sis_select
    /**
     * @brief The number of features to select during each SIS iteration
     */
    inline int n_sis_select() const {return _n_sis_select;}

    // DocString: feat_space_n_samp_train
    /**
     * @brief The nuumber of samples in the training set
     */
    inline int n_samp_train() const {return _n_samp_train;}

    // DocString: feat_space_n_feat
    /**
     * @brief The total number of features in the feature space
     */
    inline int n_feat() const {return _n_feat;}

    // DocString: feat_space_n_rung_store
    /**
     * @brief The number of rungs to calculate and store the value of the features for all samples
     */
    inline int n_rung_store() const {return _n_rung_store;}

    // DocString: feat_space_n_rung_generate
    /**
     * @brief Either 0 or 1, and is the number of rungs to generate on the fly during SIS
     */
    inline int n_rung_generate() const {return _n_rung_generate;}

    /**
     * @brief Generate a new set of non-parameterized features from a single feature
     * @details Perform all valid algebraic operations on the passed feature and all features that appear before it in _phi.
     *
     * @param feat The feature to spawn new features from
     * @param feat_set The feature set to pull features from for binary operations
     * @param start The point in feat_set to begin pulling features from for binary operations
     * @param feat_ind starting index for the next feature generated
     * @param l_bound lower bound for the maximum absolute value of the feature
     * @param u_bound upper bound for the maximum abosulte value of the feature
     */
    void generate_non_param_feats(
        std::vector<node_ptr>::iterator& feat,
        std::vector<node_ptr>& feat_set,
        const std::vector<node_ptr>::iterator& start,
        unsigned long int& feat_ind,
        const double l_bound=1e-50,
        const double u_bound=1e50
    );

    // DocString: feat_space_output_phi
    /**
     * @brief Output the feature set to a file of a passed filename
     */
    void output_phi();

#ifdef PARAMETERIZE
    /**
     * @brief Generate a new set of parameterized features from a single feature
     * @details Perform all valid algebraic operations on the passed feature and all features that appear before it in _phi.
     *
     * @param feat The feature to spawn new features from
     * @param feat_set The feature set to pull features from for binary operations
     * @param start The point in feat_set to begin pulling features from for binary operations
     * @param feat_ind starting index for the next feature generated
     * @param optimizer The object used to optimize the parameterized features
     * @param l_bound lower bound for the maximum absolute value of the feature
     * @param u_bound upper bound for the maximum abosulte value of the feature
     */
    void generate_param_feats(
        std::vector<node_ptr>::iterator& feat,
        std::vector<node_ptr>& feat_set,
        const std::vector<node_ptr>::iterator& start,
        unsigned long int& feat_ind,
        std::shared_ptr<NLOptimizer> optimizer,
        const double l_bound=1e-50,
        const double u_bound=1e50
    );

    /**
     * @brief Generate a new set of parameterized features for the residuals
     *
     * @param feat The feature to spawn new features from
     * @param feat_set The feature set to pull features from for binary operations
     * @param feat_ind starting index for the next feature generated
     * @param optimizer The object used to optimize the parameterized features
     * @param l_bound lower bound for the maximum absolute value of the feature
     * @param u_bound upper bound for the maximum abosulte value of the feature
     */
    void generate_reparam_feats(
        std::vector<node_ptr>::iterator& feat,
        std::vector<node_ptr>& feat_set,
        unsigned long int& feat_ind,
        std::shared_ptr<NLOptimizer> optimizer,
        const double l_bound=1e-50,
        const double u_bound=1e50
    );
#endif

    /**
     * @brief Generate the final rung of features on the fly and calculate their projection scores for SISat can be selected by SIS.
     *
     * @param loss The LossFunction used to project over all of the features
     * @param phi_selected The set of features that would be selected excluding the final rung
     * @param scores_selected The projection scores of all features in phi_selected
     */
    void generate_and_project(std::shared_ptr<LossFunction> loss, std::vector<node_ptr>& phi_selected, std::vector<double>& scores_selected);

    /**
     * @brief Perform Sure-Independence Screening over the FeatureSpace. The features are ranked using a projection operator constructed using _project_type and the Property vector
     *
     * @param prop Vector containing the property vector (training data only)
     */
    void sis(const std::vector<double>& prop);

    /**
     * @brief Perform Sure-Independence Screening over the FeatureSpace. The features are ranked using a projection operator defined in loss
     *
     * @param loss The LossFunction used to project over all of the features
     */
    void sis(std::shared_ptr<LossFunction> loss);

    // DocString: feat_space_feat_in_phi
    /**
     * @brief Is a feature in this process' _phi?
     *
     * @param ind (int) The index of the feature
     *
     * @return True if feature is in this rank's _phi
     */
    inline bool feat_in_phi(int ind) const {return (ind >= _phi[0]->feat_ind()) && (ind <= _phi.back()->feat_ind());}

    // DocString: feat_space_remove_feature
    /**
     * @brief Remove a feature from phi
     *
     * @param ind (int) index of feature to remove
     */
    void remove_feature(const int ind);

    #ifdef PARAMETERIZE
    // DocString: feat_space_param_feats_allowed
    /**
     * @brief True if built with -DBUILD_PARAMS (used for python tests)
     */
    bool parameterized_feats_allowed() const {return true;}
    #else
    // DocString: feat_space_param_feats_allowed
    /**
     * @brief True if built with -DBUILD_PARAMS (used for python tests)
     */
    bool parameterized_feats_allowed() const {return false;}
    #endif
    // Python Interface Functions
    #ifdef PY_BINDINGS

    // DocString: feat_space_init_file_np_array
    /**
     * @brief FeatureSpace constructor that uses a file containing postfix feature expressions to describe all features in Phi, and a primary feature setn <python/feature_creation/FeatureSpace.cpp>)
     *
     * @param feature_file (str) The file containing the postfix expressions of all features in the FeatureSpace
     * @param phi_0 (list) The set of primary features
     * @param prop (np.ndarray) List containing the property vector (training data only)
     * @param task_sizes_train (list) The number of samples in the training data per task
     * @param project_type (str) The type of loss function/projection operator to use
     * @param n_sis_select (int) The number of features to select during each SIS step
     * @param cross_corr_max (double) The maximum allowed cross-correlation value between selected features
     * @param excluded_inds (list) The list of primary feature indexes to not include in any features
     */
    FeatureSpace(
        std::string feature_file,
        py::list phi_0,
        np::ndarray prop,
        py::list task_sizes_train,
        std::string project_type="regression",
        int n_sis_select=1,
        double cross_corr_max=1.0,
        py::list excluded_inds = py::list()
    );

    // DocString: feat_space_init_file_py_list
    /**
     * @brief FeatureSpace constructor that uses a file containing postfix feature expressions to describe all features in Phi, and a primary feature setn <python/feature_creation/FeatureSpace.cpp>)
     *
     * @param feature_file (str) The file containing the postfix expressions of all features in the FeatureSpace
     * @param phi_0 (list) The set of primary features
     * @param prop (list) List containing the property vector (training data only)
     * @param task_sizes_train (list) The number of samples in the training data per task
     * @param project_type (str) The type of loss function/projection operator to use
     * @param n_sis_select (int) The number of features to select during each SIS step
     * @param cross_corr_max (double) The maximum allowed cross-correlation value between selected features
     * @param excluded_inds (list) The list of primary feature indexes to not include in any features
     */
    FeatureSpace(
        std::string feature_file,
        py::list phi_0,
        py::list prop,
        py::list task_sizes_train,
        std::string project_type="regression",
        int n_sis_select=1,
        double cross_corr_max=1.0,
        py::list excluded_inds = py::list()
    );

    // DocString: feat_space_sis_arr
    /**
     * @brief Perform Sure-Independence Screening over the FeatureSpace. The features are ranked using a projection operator constructed using _project_type and the Property vector
     *
     * @param prop (np.ndarray) Array containing the property vector (training data only)
     */
    inline void sis(np::ndarray prop)
    {
        std::vector<double> prop_vec = python_conv_utils::from_ndarray<double>(prop);
        sis(prop_vec);
    }

    // DocString: feat_space_sis_list
    /**
     * @brief Perform Sure-Independence Screening over the FeatureSpace. The features are ranked using a projection operator constructed using _project_type and the Property vector
     *
     * @param prop (list) List containing the property vector (training data only)
     */
    inline void sis(py::list prop)
    {
        std::vector<double> prop_vec = python_conv_utils::from_list<double>(prop);
        sis(prop_vec);
    }

    // DocString: feat_space_phi_selected_py
    /**
     * @brief A list containing all of the selected features
     */
    py::list phi_selected_py();

    // DocString: feat_space_phi0_py
    /**
     * @brief A list containing all features generated (Not including those created on the Fly during SIS)
     */
    py::list phi_py();

    // DocString: feat_space_phi_py
    /**
     * @brief A list containing all of the Primary features
     */
    py::list phi0_py();

    // DocString: feat_space_scores_py
    /**
     * @brief An array of all stored projection scores from SIS
     */
    inline np::ndarray scores_py(){return python_conv_utils::to_ndarray<double>(_scores);};

    // DocString: feat_space_task_sizes_train_py
    /**
     * @brief A list of the number of samples in each task for the training data
     */
    inline py::list task_sizes_train_py(){return python_conv_utils::to_list<int>(_task_sizes_train);};

    // DocString: feat_space_allowed_ops_py
    /**
     * @brief The list of allowed operators
     */
    inline py::list allowed_ops_py(){return python_conv_utils::to_list<std::string>(_allowed_ops);}

    #ifdef PARAMETERIZE
    // DocString: feat_space_allowed_ops_py
    /**
     * @brief The list of allowed operators
     */
    inline py::list allowed_param_ops_py(){return python_conv_utils::to_list<std::string>(_allowed_param_ops);}
    #else
    // DocString: feat_space_allowed_ops_py
    /**
     * @brief The list of allowed operators
     */
    inline py::list allowed_param_ops_py(){return python_conv_utils::to_list<std::string>({});}
    #endif
    // DocString: feat_space_start_rung_py
    /**
     * @brief A list containing the index of the first feature of each rung in the feature space.
     */
    inline py::list start_rung_py(){return python_conv_utils::to_list<int>(_start_rung);}

    // DocString: feat_space_get_feature
    /**
     * @brief Access the feature in _phi with an index ind
     *
     * @param ind (int) The index of the feature to get
     * @return A ModelNode of the feature at index ind
     */
    inline ModelNode get_feature(const int ind) const {return ModelNode(_phi[ind]);}
    #endif
};

#endif