Data 📈

Data is one of most important class of the library, this class store the data using by the neural network.

Declaration

This is the Data constructors.

Data(problem typeOfProblem,
     std::vector<std::vector<float>>& trainingInputs,
     std::vector<std::vector<float>>& trainingLabels,
     std::vector<std::vector<float>>& testingInputs,
     std::vector<std::vector<float>>& testingLabels,
     nature typeOfTemporal = nature::nonTemporal,
     int numberOfRecurrences = 0);

Arguments

• typeOfProblem: An enum corresponding to the type of problem associated with the data. There are 3 types of problems classification, multipleClassification and regression.
• trainingInputs: 2D vector of all the data inputs use to train the neural network. Each vector<float> represents an input for the neural network.
• trainingLabels: 2D vector of all the expected outputs use to train the neural network. Each vector<float> represents the expected output by the neural network for the corresponding input.
• testingInputs: 2D vector of all the data inputs use to evaluate the neural network. Each vector<float> represents an input for the neural network.
• testingLabels: 2D vector of all the expected outputs use to evaluate the neural network. Each vector<float> represents the expected output by the neural network for the corresponding input.
• typeOfTemporal: An enum corresponding to the temporal nature of problem associated with the data. There are 3 types of temporal nature nonTemporal, sequential and timeSeries.
• numberOfRecurrences: Size of sequence used for train neural network. Only used for timeSeries otherwise leave the value at 0.

Each example of a dataset must always be an 1D vector (except for sequential data). For example to use CIFAR-10 dataset each image must be converted to 1D vector. It’s the neural network [Input layer] which permit to define the shape of image, here Input(32, 32,3).

Data has a 2nd constructors if the data for training and testing are the same.

Data(problem typeOfProblem,
     std::vector<std::vector<float>>& inputs,
     std::vector<std::vector<float>>& labels,
     nature temporal = nature::nonTemporal,
     int numberOfRecurrences = 0);

Here is the simplest example of declaration a data for classification problem.

vector<vector<float>> inputs;
vector<vector<float>> label;
Data data(problem::classification, inputData, expectedOutputs, nature::nonTemporal);

The Data allows neuron networks to solve 3 types of problem:

enum class problem
    {
        classification,
        multipleClassification,
        regression
    };

Data can process data of many natures:

    enum class nature
    {
        nonTemporal,
        sequential,
        timeSeries,
    };

There are 3 types of problems and 3 types of temporality.