cnn_mapping_tables

CNN image-grid lookup tables for IceCube-86 and the Prometheus example.

Tensor columns use MAT_AX0_COL, MAT_AX1_COL, and MAT_AX2_COL for voxel indices. IceCube-86 is built by build_ic86_cnn_mapping(); the Prometheus example uses a fixed row list from the former prometheus_CNN_mapping.parquet layout.

graphnet.models.data_representation.images.mappings.cnn_mapping_tables.build_ic86_cnn_mapping(string_label, dom_number_label)[source]

Build the IceCube-86 CNN pixel mapping DataFrame.

Places the 78 main-array strings on a fixed 10x10 grid and folds the eight DeepCore strings (79..86) into upper/lower DeepCore views using the usual DOM numbering convention.

Parameters:

  • string_label (str) – Column name for IceCube string number (1..86).

  • dom_number_label (str) – Column name for DOM number (1..60).

Return type:

DataFrame

Returns:

DataFrame indexed by (string_label, dom_number_label) with voxel index columns mat_ax0, mat_ax1, and mat_ax2.