Correlation of Drift and Amplitude

[ekurtgl]

Hi,

I just started exploring the package and I am particularly interested in simulation of drift cases.

In the documentation it is stated that "the amplitude is modulated as a Normal distribution with amplitude 1 and standard deviation of...". I am wondering in recordings with drift whether amplitudes and stretch are independent from the amount of drift (i.e., radial distance from the electrode). Intuitively, the amplitude should reduce as a function of radial distance, and stretch should change as well, but I couldn't see this explicitly. Could you verify if that's the case in the simulation? Thank you!

Best,
Emre

[samuelgarcia]

Hi @ekurtgl
please note that spikeinterface also has generation for drifting recording
https://spikeinterface.readthedocs.io/en/stable/modules/generation.html

[ekurtgl]

Hi @samuelgarcia ,

Thank you for the information, I will check it out.

I am particularly interested in unit matching/tracking across days/sessions (something similar to UnitMatch, but with a focus on high drift cases). I am trying to figure out what would be best way of generating simulated and hybrid datasets using spikeinterface. I am thinking I can perhaps generate one drifting recording and split it into three temporal chunks, and use first and last chunks as different sessions, and throw out the second chunk (to simulate discontinuity across sessions/days).

What do you think would be the best practice here? Also, do you think it is more convenient to save the generated recording to the disk to benchmark different sorters and trackers, or should I use the lazy loading property of the module without exporting the recordings which can be long.

Also are there any example notebooks or of spikeinterface for hybrid dataset generation?

Thank you for your help!

[samuelgarcia]

When we use the same simulation sevral times (sevreal sorters/several params) we generally save it to our binary format.

about making several session : I would generate a big and unique session with an injected motion vector that have a high movement in the middle and then I would cut the sesion in 2 parts. This would mimic the inter session drift for minimal effort.

[alejoe91]

Hi @ekurtgl

In the documentation it is stated that "the amplitude is modulated as a Normal distribution with amplitude 1 and standard deviation of...". I am wondering in recordings with drift whether amplitudes and stretch are independent from the amount of drift (i.e., radial distance from the electrode). Intuitively, the amplitude should reduce as a function of radial distance, and stretch should change as well, but I couldn't see this explicitly. Could you verify if that's the case in the simulation? Thank you!

The "drift" and "modulation" are indepentent. Drift will choose which drifting template to assign to each spike based on the drifting signal. On top of this, electrode/template modulation is added as a gaussian modulation.

I am particularly interested in unit matching/tracking across days/sessions

@JoeZiminski is working on the multi-session simulation here: SpikeInterface/spikeinterface#3231

Also are there any example notebooks or of spikeinterface for hybrid dataset generation?

We have one here: https://spikeinterface.readthedocs.io/en/latest/how_to/benchmark_with_hybrid_recordings.html

[ekurtgl]

Hi @samuelgarcia and @alejoe91 ,

Thank you for the information and the resources! I will check them out.