inference

wherobots.inference

Public API for wherobots.inference.

ModelMetadata dataclass

Loads and parses Machine Learning Model Metadata JSON.

Parameters:

Name	Type	Description	Default
model_meta_path	str	Path or S3 URI to the model metadata JSON file.	required

Raises:

Type	Description
ValueError	Shape mismatch error if the length of the class_map differs from the shape of the result.
ValueError	Only MLM metadata that define statistics per band and norm_by_channels are supported.

get_mlm_properties(model_meta_uri)

Reads and parses a model metadata JSON file from S3.

Parameters:

Name	Type	Description	Default
model_meta_uri	str	The S3 URI for the model metadata JSON file.	required

Returns:

Type	Description
MLModelProperties	pystac.STACObject: The parsed model metadata conforming to the MLM Extension.

create_object_detection_udfs(batch_size, sedona)

Create and register UDFs for object detection tasks.

Registers SQL functions for RS_DETECT_BBOXES and RS_FILTER_BOX_CONFIDENCE and returns corresponding functions.

Parameters:

Name	Type	Description	Default
batch_size	int	The number of samples to include in each batch.	required
sedona	SedonaContext	The current Sedona context.	required

Returns:

Type	Description
tuple[Callable, Callable]	Tuple[Callable, Callable]: A tuple of functions configured for object detection.

create_semantic_segmentation_udfs(batch_size, sedona)

Create and register SQL functions for semantic segmentation tasks.

Registers SQL function as RS_SEGMENT and returns a corresponding function.

Parameters:

Name	Type	Description	Default
batch_size	int	The number of samples to include in each batch.	required
sedona	SedonaContext	The current Sedona context.	required

Returns:

Name	Type	Description
Callable	Callable	A function configured for semantic segmentation.

create_single_label_classification_udfs(batch_size, sedona)

Create and register SQL Functions for single-label classification tasks.

Registers SQL function as RS_CLASSIFY RS_MAX_CONFIDENCE and returns corresponding functions.

Parameters:

Name	Type	Description	Default
batch_size	int	The number of samples to include in each batch.	required
sedona	SedonaContext	The current Sedona context.	required

Returns:

Type	Description
Callable	Tuple[Callable, Callable]: A tuple of pandas UDFs configured for single-label
Callable	classification tasks, [RS_CLASSIFY, RS_MAX_CONFIDENCE].

show_detections(df, geometry_column=None, geometry_crs='EPSG:4326', confidence_threshold=0.05, plot_geoms=True, side_by_side=True)

Plot raster images with detected object geometries overlaid.

This function handles both Pandas and Wherobots DataFrames containing geometries, automatically detecting the raster column. It is compatible with dataframes returned from a SQL inference function. Exploded dataframes are not supported.

Parameters:

Name	Type	Description	Default
df	DataFrame | DataFrame	Pandas or Wherobots DataFrame containing raster data and detection results	required
geometry_column	str	Column name containing WKT geometries (if None, automatically detected)	None
geometry_crs	str	Coordinate reference system of the geometries (default: EPSG:4326)	'EPSG:4326'
confidence_threshold	float	Minimum confidence score for displaying detections	0.05
plot_geoms	bool	Whether to overlay geometries on the images	True
side_by_side	bool	Whether to show original and detection images side by side	True

Raises:

Type	Description
ValueError	If the DataFrame is empty or required columns cannot be found