Contents

class

SushiAI::Data::SyntheticClassification

Generates deterministic, linearly separable multi-class batches.

Declared in
include/SushiAI/data/dataset.hpp
Inherits
SushiAI::Data::Dataset

Samples are assigned to classes round-robin with Gaussian noise added to class means.

Public member functions

SyntheticClassification(SushiBLAS::Engine &engine, SyntheticOptions options={})

Generates the whole dataset and waits for it.

Parameters

engine

The engine that allocates and fills the tensors.

options

The problem's shape.

Exceptions

Error

If any extent in options is not positive.

virtual int64_t batch_size() const noexcept override

Samples per batch.

virtual int64_t input_dim() const noexcept override

Features per sample.

virtual int64_t num_classes() const noexcept override

How many classes the targets span.

virtual std::size_t batch_count() const noexcept override

How many batches an epoch holds.

virtual const Batch & batch(std::size_t index) const override

One batch.

Parameters

index

Which batch, in [0, batch_count()).

Returns

The batch; owned by the dataset.

Exceptions

Error

If index is out of range.

virtual std::string_view name() const noexcept override

A short name for a run report.