beamax.coefficients
Coefficient selection without materialising the full MSWPT frame.
beamax.coefficients
Memory-bounded exact top-k selection for MSGB's MSWPT analysis.
streamed_top_n_coefficients(data: jax.Array, wpt: MSWPT, *, top_n: int, input_type: str = 'spatial', positive_half: bool = False, scale: Scalar = 1.0, boxes_per_chunk: int = 8) -> tuple[jax.Array, jax.Array]
Select exact top-n scaled MSWPT coefficients without materializing them.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
data
|
(Array, shape(*N))
|
Spatial data, or an already centred Fourier array when
|
required |
wpt
|
MSWPT
|
Transform whose filters, coefficient layout, and canonical dual are used for the analysis. |
required |
top_n
|
int
|
Requested number of retained coefficients. It is clamped to the selected frame capacity, matching the configured top-n threshold. |
required |
input_type
|
('spatial', 'fourier')
|
Domain of |
"spatial"
|
positive_half
|
bool
|
Visit the first half of every level's coefficient segment rather than the complete signed frame. |
False
|
scale
|
scalar
|
Scalar applied before ranking and returning coefficient values. |
1.0
|
boxes_per_chunk
|
int
|
Static number of local Fourier boxes analysed together. |
8
|
Returns:
| Type | Description |
|---|---|
indices, values : tuple[jax.Array, jax.Array]
|
Global flat coefficient indices and scaled complex values. Rows are
ordered by ascending |
Notes
Selection state is \(\mathcal{O}(\mathtt{top\_n})\) beyond the Fourier array and one box chunk.