grid_definition

Detector-specific grid layouts: lookup tables, shapes, and scatter into tensors.

Each GridDefinition is bound to a Detector. ImageRepresentation calls GridDefinition.forward() to place pixel rows into image tensor(s).

class graphnet.models.data_representation.images.mappings.grid_definition.GridDefinition(*args, **kwargs)[source]

Bases: Model

Detector-specific orthonormal image grid(s).

Holds tensor shapes and tables that map pixel keys to voxel indices.

Construct GridDefinition.

Parameters:

  • detector (Detector) – Geometry this grid is defined for (CNN grids are detector-specific).

  • pixel_feature_names (List[str]) – Column names expected on each pixel row, including keys listed in map_pixels_by.

  • args (Any)

  • kwargs (Any)

Raises:

ValueError – If detector is not an instance of one of the detectors listed in compatible_detectors.

Return type:

object

property detector: Detector

Detector instance this grid belongs to.

abstract property compatible_detectors: List[Type[Detector]]

Detector classes this grid can be built for.

abstract property map_pixels_by: List[str]

Feature columns that join pixel rows to mappings().

abstractmethod mappings()[source]

DataFrame(s) keyed by map_pixels_by with voxel index columns.

Return type:

List[DataFrame]

abstractmethod forward(data, data_feature_names)[source]

Scatter pixel features into shaped tensors (see subclasses).

Return type:

Data

Parameters:

  • data (Data)

  • data_feature_names (List[str])

abstract property shape: List[List[int]]

Return the shape of the output images as a list of tuples.

In the dimensions (F,D,H,W) where F is the number of features per pixel. And D,H,W are the dimension of the image

class graphnet.models.data_representation.images.mappings.grid_definition.IC86GridDefinition(*args, **kwargs)[source]

Bases: GridDefinition

IceCube-86 CNN grid (main array + DeepCore) layouts and lookup table.

Construct IC86GridDefinition.

Parameters:

  • detector (Detector) – IceCube86 instance (grid is fixed to that geometry).

  • dtype (dtype) – data type used for node features. e.g. ´torch.float´

  • string_label (str, default: 'string') – Name of the feature corresponding to the DOM string number. Values Integers between 1 - 86

  • dom_number_label (str, default: 'dom_number') – Name of the feature corresponding to the DOM number (1 - 60). Values Integers between 1 - 60 where 1 is the dom with the highest z coordinate.

  • pixel_feature_names (List[str]) – Names of each column in expected input data that will be built into a image.

  • include_main_array (bool, default: True) – If True, the main array will be included.

  • include_lower_dc (bool, default: True) – If True, the lower DeepCore will be included.

  • include_upper_dc (bool, default: True) – If True, the upper DeepCore will be included.

  • args (Any)

  • kwargs (Any)

Raises:

ValueError – If no array type is included.

Return type:

object

NOTE: Expects input data to be DOMs with aggregated features.

property compatible_detectors: List[Type[Detector]]

Grid is fixed to the IceCube-86 geometry.

property map_pixels_by: List[str]

String and DOM identifiers used for voxel lookup.

mappings()[source]

Return the single combined lookup table for all sub-images.

Return type:

List[DataFrame]

forward(data, data_feature_names)[source]

Scatter pixel rows into IceCube-86 image tensor(s).

Return type:

Data

Parameters:

  • data (Data)

  • data_feature_names (List[str])

property shape: List[List[int]]

Return the shape of the output images as a list of tuples.

class graphnet.models.data_representation.images.mappings.grid_definition.ExamplePrometheusGridDefinition(*args, **kwargs)[source]

Bases: GridDefinition

Example single-image grid for Prometheus-style layouts.

Construct grid.

Parameters:

  • detector (Detector) – Typically ORCA150 in the example scripts.

  • dtype (dtype) – data type used for node features. e.g. ´torch.float´

  • string_label (str, default: 'sensor_string_id') – Name of the feature corresponding to the sensor string number.

  • sensor_number_label (str, default: 'sensor_id') – Name of the feature corresponding to the sensor number

  • pixel_feature_names (List[str]) – Names of each column in expected input data that will be built into a image.

  • args (Any)

  • kwargs (Any)

Raises:

ValueError – If no array type is included.

Return type:

object

NOTE: Expects input data to be sensors with aggregated features.

property compatible_detectors: List[Type[Detector]]

Example grid is built for the ORCA150 geometry.

property map_pixels_by: List[str]

String and sensor identifiers used for voxel lookup.

mappings()[source]

Return the embedded lookup table for the example layout.

Return type:

List[DataFrame]

forward(data, data_feature_names)[source]

Scatter pixel rows into the example 3D image tensor.

Return type:

Data

Parameters:

  • data (Data)

  • data_feature_names (List[str])

property shape: List[List[int]]

Return the shape of the output images as a list of tuples.