search

flatironinstitute / ironclust

Spike sorting software being developed at Flatiron Institute, based on JRCLUST (Janelia Rocket Cluster)
Apache License 2.0
28 stars 7 forks source link

Traces #40

Closed lorenzfenk closed 4 years ago

lorenzfenk commented 4 years ago

Hello, I am quite new to spike-sorting more generally and using ironclust in particular, repeatedly encountering the following issue when trying to sort my Neuropixels (3B staggered) recordings using ironclust (v5.0.5 (10/14/2019)): Clusters seem ok, peak amplitudes are detected at the right channels, and spike waveforms also look good (i.e. what to expect from inspecting the raw data). However, when I look at the traces (>> irc traces...) it is almost always regions outside the actual spikes that get labelled (and stored with the corresponding time stamps). In fact, even if I set the very threshold high, it is not big spikes that get labelled but these small wiggles instead. I have attached an example for demonstration (qqFactor=5). In the overview you can see how the vast majority of spikes is missed in traces. In the second example – zooming in on channel 21 – not a single spike was correctly labelled in ‘traces’ over a period of 2min, yet the overlaid waveforms of ~100 spikes look just fine (and any wiggles/flat traces are absent)... I imagine if that was due to a bug it would have been noticed by many others before, so maybe doing something stupid, but would be grateful for your advice. all the best! lorenz

Screen Shot 2020-02-03 at 20 49 48 Screen Shot 2020-02-02 at 16 19 20
jamesjun commented 4 years ago

Thanks for reporting the error. I implemented irc2 traces [myparam.prm] command. Please see README.md for more details.

lorenzfenk commented 4 years ago

Dear James, Thanks for your reply, I will test irc2 asap. However, regarding the problem I have been having when running irc traces, I have followed up on that: The issue arises only when specifying a time range in “tlim_load”; ie if the field is empty or the time window starts with 0 it is all fine, but as soon as the time range doesn’t start with zero the problem occurs (e.g. [0 1200] will work, [10 1200] won’t). Attached two screen shots for visualization and after analyzing a short data stretch with otherwise identical parameters, yet specifying tlim_load = [0 60] in the first (upper example) and [1 60] in the second (below). best, lorenz

0 1