training-data-manager.h

Go to the documentation of this file.

#pragma once

#include <tuple>
#include <vector>

#include <GRT/GRT.h>

using std::vector;

class TrainingDataManager {
  public:
    // Constructor
    TrainingDataManager(uint32_t num_classes);

    // Set the dimension of the training data
    bool setNumDimensions(uint32_t dim);

    // Set the name of the training data
    bool setDatasetName(const std::string name);

    // Set the name of the training data
    bool setDatasetName(const char* const name);

    GRT::TimeSeriesClassificationData getAllData() { return data_; }

    uint32_t getNumLabels() { return num_classes_; }

    uint32_t getTotalNumSamples() { return data_.getNumSamples(); }

    // =================================================
    //  Functions that enables per-sample naming
    // =================================================

    bool setNameForLabel(const std::string name, uint32_t label);
    std::string getLabelName(uint32_t label);

    std::string getSampleName(uint32_t label, uint32_t index);
    bool setSampleName(uint32_t, uint32_t, const std::string);

    // =================================================
    //  Functions that simplifies editing
    // =================================================

    bool addSample(uint32_t label, const GRT::MatrixDouble& sample);

    uint32_t getNumSampleForLabel(uint32_t label);

    GRT::MatrixDouble getSample(uint32_t label, uint32_t index);

    bool deleteSample(uint32_t label, uint32_t index);

    bool deleteAllSamples();

    bool deleteAllSamplesWithLabel(uint32_t label);

    bool relabelSample(uint32_t label, uint32_t index, uint32_t new_label);

    bool trimSample(uint32_t label, uint32_t index, uint32_t start,
                    uint32_t end);

    // =================================================
    //  Functions that manage per-sample scores
    // =================================================

    bool hasSampleScore(uint32_t label, uint32_t index);
    double getSampleScore(uint32_t label, uint32_t index);
    bool setSampleScore(uint32_t label, uint32_t index, double score);

    bool hasSampleClassLikelihoods(uint32_t label, uint32_t index);
    vector<double> getSampleClassLikelihoods(uint32_t label, uint32_t index);
    bool setSampleClassLikelihoods(uint32_t label, uint32_t index,
                                   vector<double> likelihoods);

    // =================================================
    //  Functions for saving/loading training data
    // =================================================

    inline bool save(const std::string& filename) {
        return data_.save(filename);
    }

    bool load(const std::string& filename);

  private:
    uint32_t num_classes_;

    // Name simulates Option<std::string> type. If `Name.first` is true, then
    // the name is valid; else use the default name.
    using Name = std::pair<bool, std::string>;
    vector<vector<Name>> training_sample_names_;
    vector<std::string> default_label_names_;

    // Score simulates Option<double> type. If `Score.first` is true, then
    // the score is valid; else use the default score.
    using Score = std::pair<bool, double>;
    vector<vector<Score>> training_sample_scores_;

    // Probability that the sample belongs to each class (e.g. when the model
    // is trained on all other samples). If ClassLikelihoods.first is true,
    // then vector is valid.
    using ClassLikelihoods = std::pair<bool, vector<double>>;
    vector<vector<ClassLikelihoods>> training_sample_class_likelihoods_;

    // This variable tracks the number of samples for each label. Although We
    // can get the number with TimeSeriesClassificationData::getClassData and
    // then getNumSamples. Caching the information here helps with bound checks!
    //
    // The use of this class requires index not beyond the
    // num_samples_per_label_.
    vector<uint32_t> num_samples_per_label_;

    // The underlying data store backed up by GRT's TimeSeriesClassificationData
    GRT::TimeSeriesClassificationData data_;

    // Disallow copy and assign
    TrainingDataManager(TrainingDataManager&) = delete;
    void operator=(TrainingDataManager) = delete;
};