Unit classifier for AIND ephys pipeline

aind-ephys-unit-classifier

Description

This capsule is designed to automatically label spike sorted units for the AIND pipeline.

It uses pre-trained models to classify units as:

• noise (non-neuronal)
• MUA (multi-unit activity)
• SUA (single-unit activity)

The model was developed by the Musall's Group at the Forschungszentrum Jülich, Germany.

The model files are in the code/unit_classifier_models_v1.0 folder.

• metrics.json: list of required metrics for the models
• noise-neuron_classifier.pkl: the scikit-learn model for noise VS neuron classification
• sua-mua_classifier.pkl: the scikit-learn model for sua VS mua classification

Inputs

The data/ folder must include the output of the aind-ephys-postprocessing, including the postprocessed_{recording_name} folder.

Parameters

The code/run script takes no arguments.

Output

The output of this capsule is the following:

• results/unit_classifier_{recording_name}.csv file, containing the decoder_labels and decoder_probability for each unit
• results/data_process_unit_classifier_{recording_name}.json file, a JSON file containing a DataProcess object from the aind-data-schema package.