MemoryError after changing sampling frequency

[IljaPilja]

MemoryError after changing sampling frequency

[edeno]

Hey,

So you're running out of memory, probably because of how highly sampled your data is. I would downsample your data to a more manageable sampling rate. For most things, if you downsample to 1000 Hz (for spikes and LFPs), that should still give you enough resolution to resolve frequencies up to 500 Hz.

[IljaPilja]

Hi,
so when I resample my 300 ms data piece (aka 6000 data ticks) at 20 kHz to 1 kHz, the LFP looks ok, but my binary vector containing ones for my spike times only consists of zeros since the resampling step has removed them all. Thus the coherence measurement does not work with spike vectors consisting of only zeros. Besides the factor of 20 downsampling I tried 10, 5, 2.5, the latter produces the MemoryError again and the others give me an almost flat coherence signal. Guess my data does not work with this function. Thanks for the quick response though.

[edeno]

There's nothing in downsampling that will remove your spike times.

For example if you spike times are 0.001, 0.005, 0.007 seconds and you had 1 ms bins (sampling rate: 500 Hz), you would have something like 1, 0, 0, 0, 1, 0, 1.

If you downsample to 2 ms bins (sampling rate: 1000 Hz), this would now be: 1, 0, 1, 1

[Mark-Kramer]

For the missing spike times, an option could be:

1. Get the spike times at high sampling frequency.
2. Downsample the LFP to 1 kHz.
3. Make a vector of zeros, with sampling rate 1 kHz, and length equal to that of the LFP.
4. Put this spike times from (1) at the correct temporal locations in the vector in (3).

This would perhaps, preserve your spike times in (1), and let you downsample in (2).

[IljaPilja]

Thanks for the great help. I'm using the rebin package from Python to perform the downsampling and this works great with my data and I'm able to now determine the coherence.
Thank you again for your help.