Skip to content

torchsky.quality

Stability: core. Names on this page follow the deprecation policy in contributing.

Torch-native quality, inverse-variance, and confidence contracts.

Guides: coadds

This reference is generated from the package source. Names are grouped by the module that defines them; import every name from the public namespace torchsky.quality unless the group text says otherwise.

Top-level namespace

Torch-native quality, inverse-variance, and confidence contracts.

CONFIDENCE_SEMANTICS module-attribute 

CONFIDENCE_SEMANTICS = 'normalized_weight_0_to_1'

CONTRACT_VERSION module-attribute 

CONTRACT_VERSION = 'torchsky.quality.v1'

DEFAULT_ZERO_WEIGHT_BITS module-attribute 

DEFAULT_ZERO_WEIGHT_BITS = (
    BAD_PIXEL
    | SATURATED
    | COSMIC_RAY
    | STREAK
    | INVALID_VARIANCE
)

INVERSE_VARIANCE_SEMANTICS module-attribute 

INVERSE_VARIANCE_SEMANTICS = 'inverse_variance_per_pixel'

MASK_POLARITY module-attribute 

MASK_POLARITY = 'set_means_flagged'

QualityBit

Bases: IntFlag

Named quality conditions stored as set bits in an integer mask.

BAD_PIXEL class-attribute instance-attribute 

BAD_PIXEL = 1 << 0

COSMIC_RAY class-attribute instance-attribute 

COSMIC_RAY = 1 << 2

DETECTED class-attribute instance-attribute 

DETECTED = 1 << 3

INVALID_VARIANCE class-attribute instance-attribute 

INVALID_VARIANCE = 1 << 5

SATURATED class-attribute instance-attribute 

SATURATED = 1 << 1

STREAK class-attribute instance-attribute 

STREAK = 1 << 4

WeightProduct dataclass 

WeightProduct(
    quality_mask: Tensor,
    inverse_variance: Tensor,
    weight: Tensor,
    confidence: Tensor,
    metadata: Mapping[str, Any],
)

Classical quality products with reserved producer metadata.

confidence instance-attribute 

confidence: Tensor

inverse_variance instance-attribute 

inverse_variance: Tensor

metadata instance-attribute 

metadata: Mapping[str, Any]

quality_mask instance-attribute 

quality_mask: Tensor

weight instance-attribute 

weight: Tensor

build_weight_product 

build_weight_product(
    inverse_variance: Tensor,
    quality_mask: Tensor | None = None,
    *,
    exclude_detected: bool = False,
    confidence_percentile: float = 99.0,
    producer: Mapping[str, Any] | None = None,
) -> WeightProduct

Sanitize inverse variance and combine it with named quality bits.

Invalid variance is represented by zero inverse variance and zero weight, and is recorded as INVALID_VARIANCE. producer reserves a stable artifact field for future learned mask/weight generators without changing downstream consumers.

Parameters:

Name Type Description Default
inverse_variance Tensor

Per-pixel inverse variance; nonpositive or nonfinite values are invalid.

 required
quality_mask Tensor | None

Optional integer bit mask with the same shape.

 None
exclude_detected bool

Give detected-source pixels zero weight when true.

 False
confidence_percentile float

Positive-weight percentile mapped to confidence one.

 99.0
producer Mapping[str, Any] | None

Optional provenance mapping stored in the result metadata.

 None

Returns:

Type Description
WeightProduct

Sanitized mask, inverse variance, weight, confidence, and metadata.

Examples:

>>> import torch
>>> from torchsky.quality import build_weight_product
>>> build_weight_product(torch.tensor([1.0])).weight.tolist()
[1.0]

quality_bit_names 

quality_bit_names(
    value: int | QualityBit,
) -> frozenset[str]

Decode a bitset into stable public names.

quality_bits_from_names 

quality_bits_from_names(
    names: Mapping[str, Any]
    | list[str]
    | tuple[str, ...]
    | set[str],
) -> QualityBit

Encode stable public quality names into a bitset.