Improve template matching in SpikeInterface
Key Investigators
Pierre Yger Samuel Garcia Joulien Boussard
Project Description
Objectives
Many spike sorters use template matching aka peeler aka deconvolution as a final step in the spike sorting pipeline. It is one of the best option to handle the spike collision issue.
There are many algorithms that can be used for that steps.
SpikeInterface already includes 3 of then:
- tridesclous strategy
- spyking-cicus strategy
- cicus-omp (orthogonal matching pursuit) strategy
These 3 methods give decent results, but bencmarks are still limited.
The hackathon would be an oportunity to improve the existing strategies and port more methods for a more systematic and complete benchmark.
Approach and Plan
Implement methods from other sorter:
- YASS
- Kilosort
- HDSort
Make notebooks to report benchmarks in various conditions:
- low/high channel count
- stationary vs drift dataset
- low/high spike collisions
- low/high bursting modulation