Contents

struct

SushiAI::Data::SyntheticOptions

How to shape the generated classification problem.

Declared in
include/SushiAI/data/dataset.hpp

Public attributes

int64_t batch_size = 32

Samples per batch.

int64_t batches = 16

How many batches to generate.

int64_t features = 16

Features per sample.

int64_t classes = 4

How many classes.

double noise = 0.35

Noise standard deviation, relative to unit class means.

Below about 0.7 the classes stay comfortably separable, so a correct trainer reaches near-perfect accuracy and a test can assert it.

std::uint64_t seed = 0x0D15EA5E'0BADF00Dull

The Philox seed; the same seed gives the same data.