RotatedDigitsMetaDataset<T, TInput, TOutput>
Models & Types · AiDotNet.MetaLearning.Data
A synthetic meta-dataset for image-like classification where each class is a rotated “digit” pattern (a simple 2D feature vector derived from an angle). Different tasks use different subsets of rotation angles, simulating few-shot image classification benchmarks.
Inheritance: MetaDatasetBase<T, TInput, TOutput> → RotatedDigitsMetaDataset<T, TInput, TOutput>
For Beginners
This creates a simplified version of the Omniglot-style benchmark. Each “digit” is represented by a rotation angle that produces a 2D feature vector via (cos(angle), sin(angle)) plus noise. The meta-learner must learn to classify which rotation group a new point belongs to from just a few examples.
Constructors
| Constructor | Summary |
|---|---|
RotatedDigitsMetaDataset(int numClasses, int examplesPerClass, int featureDim, double noiseStdDev, int? seed) |
Creates a rotated digits meta-dataset. |
Properties
| Property | Summary |
|---|---|
IReadOnlyDictionary<int, int> ClassExampleCounts { get; } |
|
int TotalClasses { get; } |
|
int TotalExamples { get; } |
|
string Name { get; } |