features

Documentation

Index

• type SpectralFingerprint
  • func NewSpectralFingerprint[T tensor.Numeric](engine compute.Engine[T], ops numeric.Arithmetic[T], topK int) *SpectralFingerprint[T]
  • func (s *SpectralFingerprint[T]) Attributes() map[string]interface{}
  • func (s *SpectralFingerprint[T]) Backward(_ context.Context, _ types.BackwardMode, _ *tensor.TensorNumeric[T], ...) ([]*tensor.TensorNumeric[T], error)
  • func (s *SpectralFingerprint[T]) Forward(ctx context.Context, inputs ...*tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
  • func (s *SpectralFingerprint[T]) OpType() string
  • func (s *SpectralFingerprint[T]) OutputShape() []int
  • func (s *SpectralFingerprint[T]) Parameters() []*graph.Parameter[T]

Types

type SpectralFingerprint

type SpectralFingerprint[T tensor.Numeric] struct {
	TopK int
	// contains filtered or unexported fields
}

SpectralFingerprint is a feature transformation layer that computes the DFT of a time-series input and returns the magnitudes of the top K frequencies. This layer is non-trainable and is typically used for feature extraction. Because it is non-trainable, it does not propagate gradients through this transformation (Backward returns nil), effectively treating it as a fixed feature extractor.

Computation uses engine primitives (MatMul, Mul, Add, Sqrt) with precomputed DFT basis matrices, making the layer fully traceable by the tracing compiler.

func NewSpectralFingerprint

func NewSpectralFingerprint[T tensor.Numeric](engine compute.Engine[T], ops numeric.Arithmetic[T], topK int) *SpectralFingerprint[T]

NewSpectralFingerprint creates a new SpectralFingerprint layer.

func (*SpectralFingerprint[T]) Attributes

func (s *SpectralFingerprint[T]) Attributes() map[string]interface{}

Attributes returns the attributes of the layer.

func (*SpectralFingerprint[T]) Backward

func (s *SpectralFingerprint[T]) Backward(_ context.Context, _ types.BackwardMode, _ *tensor.TensorNumeric[T], inputs ...*tensor.TensorNumeric[T]) ([]*tensor.TensorNumeric[T], error)

Backward for a non-trainable layer returns a nil gradient for the input.

func (*SpectralFingerprint[T]) Forward

func (s *SpectralFingerprint[T]) Forward(ctx context.Context, inputs ...*tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

Forward computes the forward pass of the layer.

func (*SpectralFingerprint[T]) OpType

func (s *SpectralFingerprint[T]) OpType() string

OpType returns the operation type of the layer.

func (*SpectralFingerprint[T]) OutputShape

func (s *SpectralFingerprint[T]) OutputShape() []int

OutputShape returns the output shape of the layer.

func (*SpectralFingerprint[T]) Parameters

func (s *SpectralFingerprint[T]) Parameters() []*graph.Parameter[T]

Parameters returns no parameters as this layer is not trainable.