IEpisodicDataset<T, TInput, TOutput>

Interfaces · AiDotNet.MetaLearning.Data

Represents a dataset that can sample episodic tasks for meta-learning.

For Beginners

An episodic dataset is a special kind of dataset used in meta-learning. Instead of just giving you examples one at a time, it can create complete “tasks” or “episodes” that each have their own training and test sets.

For example, if you have a dataset of animal images, an episodic dataset can:

  1. Randomly pick 5 animal types (e.g., cat, dog, bird, fish, rabbit) - this is “5-way”
  2. Give you 1 image of each animal for training - this is “1-shot”
  3. Give you more images of those same animals for testing

This allows the model to practice learning new tasks quickly, which is the core idea of meta-learning.

Properties

Property Summary
DatasetSplit Split { get; } Gets the split type of this dataset (train, validation, or test).
Dictionary<int, int> ClassCounts { get; } Gets the number of examples per class in the dataset.
int NumClasses { get; } Gets the total number of classes available in the dataset.

Methods

Method Summary
IMetaLearningTask<T, TInput, TOutput>[] SampleTasks(int numTasks, int numWays, int numShots, int numQueryPerClass) Samples a batch of N-way K-shot tasks from the dataset.
void SetRandomSeed(int seed) Sets the random seed for reproducible task sampling.