IMetaDataset<T, TInput, TOutput>

Interfaces · AiDotNet.Interfaces

Represents a high-level meta-dataset that can generate episodes for meta-learning. Unlike IEpisodicDataset which operates on pre-built episodes, this interface generates episodes on-the-fly from an underlying data source.

Known implementations: MetaDatasetBase<T, TInput, TOutput>, MetaDatasetFormat<T, TInput, TOutput>

For Beginners

A meta-dataset is a collection of data organized so that we can create many different learning tasks from it. Each task (episode) contains a small training set (support) and a small test set (query). The meta-learner trains across many such tasks to learn how to learn quickly from small amounts of data.

Properties

Property Summary
IReadOnlyDictionary<int, int> ClassExampleCounts { get; } Gets the number of examples available for each class, keyed by class index.
int TotalClasses { get; } Gets the total number of distinct classes available in the dataset.
int TotalExamples { get; } Gets the total number of examples across all classes.
string Name { get; } Gets the name of this meta-dataset.

Methods

Method Summary
IEpisode<T, TInput, TOutput> SampleEpisode(int numWays, int numShots, int numQueryPerClass) Samples a single episode from the dataset.
IReadOnlyList<IEpisode<T, TInput, TOutput>> SampleEpisodes(int count, int numWays, int numShots, int numQueryPerClass) Samples multiple episodes from the dataset.
bool SupportsConfiguration(int numWays, int numShots, int numQueryPerClass) Gets whether this dataset supports a given N-way K-shot configuration.
void SetSeed(int seed) Sets the random seed for reproducible episode generation.