Interface IEpisodicDataLoader<T, TInput, TOutput>

Namespace

AiDotNet.Interfaces

Assembly

AiDotNet.dll

Interface for data loaders that provide episodic tasks for meta-learning.

public interface IEpisodicDataLoader<T, TInput, TOutput> : IDataLoader<T>, IResettable, ICountable, IBatchIterable<MetaLearningTask<T, TInput, TOutput>>

Type Parameters

T

The numeric type used for calculations, typically float or double.

TInput

The input data type for tasks.

TOutput

The output data type for tasks.

Inherited Members

IDataLoader<T>.Name

IDataLoader<T>.Description

IDataLoader<T>.IsLoaded

IDataLoader<T>.LoadAsync(CancellationToken)

IDataLoader<T>.Unload()

IResettable.Reset()

ICountable.TotalCount

ICountable.CurrentIndex

ICountable.BatchCount

ICountable.CurrentBatchIndex

ICountable.Progress

IBatchIterable<MetaLearningTask<T, TInput, TOutput>>.BatchSize

IBatchIterable<MetaLearningTask<T, TInput, TOutput>>.HasNext

IBatchIterable<MetaLearningTask<T, TInput, TOutput>>.GetNextBatch()

IBatchIterable<MetaLearningTask<T, TInput, TOutput>>.TryGetNextBatch(out MetaLearningTask<T, TInput, TOutput>)

IBatchIterable<MetaLearningTask<T, TInput, TOutput>>.GetBatches(int?, bool, bool, int?)

IBatchIterable<MetaLearningTask<T, TInput, TOutput>>.GetBatchesAsync(int?, bool, bool, int?, int, CancellationToken)

Extension Methods

ParallelBatchLoaderExtensions.WithParallelLoading<TBatch>(IBatchIterable<TBatch>, Func<int[], TBatch>, int, int?, int?)

DataPipelineExtensions.ToPipeline<TBatch>(IBatchIterable<TBatch>)

DataLoaderExtensions.CreateBatchesAsync<TBatch>(IBatchIterable<TBatch>, int?, int)

DataLoaderExtensions.CreateBatches<TBatch>(IBatchIterable<TBatch>, int?)

Remarks

This interface is for meta-learning scenarios using N-way K-shot learning, where the loader generates tasks consisting of: - Support set: K examples per class for N classes (used to adapt the model) - Query set: Additional examples for evaluation after adaptation

For Beginners: Meta-learning is "learning to learn".

Standard ML: Train on lots of cat/dog images, then classify new cat/dog images.

Meta-learning: Train on many different tasks (cats vs dogs, cars vs planes, etc.), then when given a new task with only a few examples, quickly learn to do it.

N-way K-shot means:

• N-way: Each task has N different classes to distinguish
• K-shot: You get K examples of each class to learn from

Example: 5-way 1-shot

• Given 5 new animal types you've never seen
• With only 1 example image of each
• Classify new images into one of these 5 types

The episodic data loader creates these mini-tasks for training.

Properties

AvailableClasses

Gets the total number of available classes in the dataset.

int AvailableClasses { get; }

Property Value

int

KShot

Gets the number of support examples per class (K in K-shot).

int KShot { get; }

Property Value

int

NWay

Gets the number of classes per task (N in N-way).

int NWay { get; }

Property Value

int

QueryShots

Gets the number of query examples per class.

int QueryShots { get; }

Property Value

int

Methods

GetNextTask()

Gets the next meta-learning task (support set + query set).

MetaLearningTask<T, TInput, TOutput> GetNextTask()

Returns

MetaLearningTask<T, TInput, TOutput>

A MetaLearningTask with support and query sets.

Remarks

Each call returns a new randomly sampled task with: - N randomly selected classes from available classes - K support examples per class - QueryShots query examples per class

GetTaskBatch(int)

Gets multiple meta-learning tasks as a batch.

IReadOnlyList<MetaLearningTask<T, TInput, TOutput>> GetTaskBatch(int numTasks)

Parameters

numTasks int

Number of tasks to sample.

Returns

IReadOnlyList<MetaLearningTask<T, TInput, TOutput>>

A list of MetaLearningTasks.

SetSeed(int)

Sets the random seed for reproducible task sampling.

void SetSeed(int seed)

Parameters

seed int

Random seed value.