coco_pipe.descriptors#

Descriptor extraction and the table lifecycle.

The descriptor lifecycle has four stages:

  1. extract — DescriptorPipeline turns (n_obs, n_channels, n_times) arrays into a flat DataContainer of descriptors.

  2. reject — epoch MAD-outlier rejection via coco_pipe.io.quality.drop_epoch_outliers() (+ mad_failures_from_qc() for failure logging).

  3. aggregate — build_descriptor_tables() builds per-epoch and group-aggregated subject tables (+ add_aggregated_band_ratios()).

  4. merge — merge_descriptor_tables() concatenates per-shard tables of one table kind; save_descriptor_table() / load_descriptor_table() handle on-disk descriptor tables.

Submodules#