Aurélien WYNGAARD

I am curious about this, is it still to be implemented?

Same, this option would be very welcome!

I think I have a pretty good idea of where the problem is: I made a random spike train for 3 units and plotted their auto and cross-correlograms, here is...

Ah good catch! I think we can easily remedy the situation. We need to replace `range(i, ...)` by `range(...)` and remove the flip here: https://github.com/SpikeInterface/spikeinterface/blob/99003fc1cf99ce322ad7eaff5847a9ab2f4477f9/spikeinterface/postprocessing/correlograms.py#L403 This will make the computation...

Shouldn't `method="numpy"` and `method="numba"` have the exact same output? I think it can be fixed with what I wrote a week ago no?

I guess then the only fix would be to compute all the correlograms using the spike vector, but that would need an entire rewrite and we would lose parallelization. Probably...

> ok, should I merge as is? I'm good with it!

```python amplitudes = st.compute_spike_amplitudes(wvf_extractor, load_if_exists=False, chunk_duration='2s', n_jobs=12) spikeinterface.exporters.export_to_phy(wvf_extractor, "/mnt/raid0/data/MEArec/kilosort2_default_phy", compute_pc_features=False, compute_amplitudes=True, copy_binary=False, max_channels_per_template=24, template_mode="average", chunk_duration='2s', n_jobs=8) ``` Works fine ... except that I cannot open the folder with `phy template-gui...

`amplitude_segment_0.npy` has a dtype of `float32`. I ran `compute_spike_amplitudes(return_scaled=True)` but `extract_waveforms(return_scaled=False)` Does it come from here?

> This is strange that amplitude_segment_0.npy has a dtype of float32 and the one copied in phy is int64. No the one in phy is `float32`. Is it `spike_templates.npy` that...