Question: Is possible/reasonable to include "select_peaks" to run_node_pipeline?

[borrepp]

Hello,

During our preprocessing pipeline, we are plotting the peak locations. We are in the process of updating SpikeInterface to version 0.104.8, and I saw that now there is a new way to improve the speed of this process: detect_peaks -> localize_peaks (https://spikeinterface.readthedocs.io/en/0.104.8/modules/sortingcomponents.html#node-pipelines).

We want to implement this new feature. However, when plotting peak locations, it is always useful to plot a subsample of the peaks (via "select_peaks"); otherwise, it can be difficult to appreciate motion.

I was just wondering if "select_peaks" can also be added to the node_pipeline. If so, how can it be implemented? Or, given the efficiency of using run_node_pipeline, should we do the subsample after we get the peaks and peaks_locations?

Thanks in advance for your help¡
Best
Pepe

[alejoe91]

Hi @borrepp

I recommend computing all peaks, but using the sw.plot_drift_raster_map function. The scatter_decimate argument does exactly what you're asking for :)

[borrepp]

Hi @alejoe91
Thank you so much. I will follow the recommendation.
Thanks.

[borrepp]

Hi @alejoe91

Just wanted to mention that I computed all peaks and peak_locations as you suggested, but instead of using the scatter_decimate argument for the plotting, I used "select_peaks" to create "peaks_sampled" using "smart_sampling_amplitudes"; that function also returned the indices that can be used to sample the peak_locations. Then I use peaks_sampled & peaks_locations[sampled_indices] with sw.plot_drift_raster_map function and scatter_decimate = None

I think this option can leverage the algorithms to sample the peaks better. What do you think?

Thanks for the tips¡

By the way, the peak location processing runs much faster with the run_node_pipeline ¡¡ Thanks again¡
Best
Pepe