dataset

Dataset Class Documentation

The Dataset class is a C++ class that represents a dataset of digit images for training and testing purposes in MNIST digit recognition. This documentation provides a detailed overview of the class structure, functionality, implementation details, and conclusion.

Class Structure

The Dataset class has the following structure:

class Dataset {
private:
    static Dataset *instance;
    static std::mutex mutex_;

protected:
    vector<DigitImage> train_data;
    vector<DigitImage> test_data;
    std::vector<size_t> unique_indices;

    Dataset();
    ~Dataset();

public:
    size_t training_data_size;
    size_t test_data_size;

    Dataset(Dataset& other) = delete;
    void operator=(const Dataset&) = delete;

    std::vector<DigitImage> extract_training_batch(size_t batch_size, int seed = RANDOM_SEED);
    std::vector<DigitImage> get_training_data();
    std::vector<DigitImage> get_test_data();

    static Dataset* GetInstance();
};

classDiagram
class Dataset {
- static Dataset *instance
- static std::mutex mutex_
- vector<DigitImage> train_data
- vector<DigitImage> test_data
- std::vector<size_t> unique_indices
- size_t training_data_size
- size_t test_data_size
+ Dataset()
+ ~Dataset()
+ std::vector<DigitImage> extract_training_batch(size_t batch_size, int seed = RANDOM_SEED)
+ std::vector<DigitImage> get_training_data()
+ std::vector<DigitImage> get_test_data()
+ static Dataset *GetInstance()
}

Loading

Functionality

The Dataset class provides the following functionality:

• Private members:

  • instance: A pointer to the singleton instance of the Dataset class.
  • mutex_: A mutex for thread-safe initialization of the singleton instance.
  • train_data: A vector of DigitImage objects representing the training data.
  • test_data: A vector of DigitImage objects representing the test data.
  • unique_indices: A vector of indices used for random selection of training data.

• Public members:

  • training_data_size: The size of the training data.
  • test_data_size: The size of the test data.

• Constructors and Destructors:

  • Dataset(): Private constructor that loads the MNIST dataset and initializes the training and test data.
  • ~Dataset(): Default destructor.

• Deleted Members:

  • Dataset(Dataset& other): Copy constructor is deleted to prevent copying of the singleton instance.
  • operator=(const Dataset&): Copy assignment operator is deleted to prevent copying of the singleton instance.

• Member Functions:

  • std::vector<DigitImage> extract_training_batch(size_t batch_size, int seed = RANDOM_SEED): Extracts a batch of DigitImage objects from the training data. The batch size and an optional seed for random shuffling can be specified.
  • std::vector<DigitImage> get_training_data(): Retrieves the training data as a vector of DigitImage objects.
  • std::vector<DigitImage> get_test_data(): Retrieves the test data as a vector of DigitImage objects.
  • static Dataset* GetInstance(): Retrieves the singleton instance of the Dataset class.

Implementation Details

• The Dataset class follows the singleton design pattern to ensure that only one instance of the class exists.
• The class uses a private constructor to load the MNIST dataset and initialize the training and test data.
• The train_data and test_data vectors store DigitImage objects representing the digit images in the dataset.
• The unique_indices vector is used to keep track of indices for random selection of training data.
• The class provides member functions to extract training batches, retrieve the training and test data, and retrieve the singleton instance.
• The singleton instance is lazily initialized and accessed using the GetInstance() static member function, which ensures thread safety using a mutex.

Conclusion

The Dataset class provides a convenient interface for accessing the MNIST dataset for digit recognition tasks. It manages the training and test data, provides methods for extracting training batches,

and follows the singleton design pattern to ensure a single instance of the dataset is used throughout the program.