NumpySorting.from_unit_dict() spike samples instead of times

[pas-calc]

The function NumpySorting.from_unit_dict sais

Each dict have unit_ids as keys and spike times as values

but units_dict_list is interpreted to have values as samples

spikeinterface/src/spikeinterface/core/numpyextractors.py

Line 410 in 25940b4

sample_indices.append(spike_times)

Because of minimum_spike_dtype the data is casted from seconds float to int64

Either we adjust the docstring, or it could be nice to have a convenience function like from_unit_dict where we can define whether to be interpreted as samples or seconds (like the recent updates to spikeinterface clarified NumpySorting.from_times_and_labels / NumpySorting.from_samples_and_labels we could name them differently )

Edit: Actually I was looking for a function to create a sorting from a Pynapple TsGroup (in seconds) as to_pynapple_tsgroup creates, so to read that again without having access to the original Sorting/SortingAnalyzer.

# spike_trains = dict(session.units)
spike_trains = {unit_id: st.t for unit_id,st in units_tsgroup.items()} # units_tsgroup is a nap.TsGroup , convert times from nap.Ts (timestamp) to numpy array
spike_trains.keys() # unit ids
sorting = NumpySorting.from_unit_dict(spike_trains, sampling_frequency=sampling_frequency)

[pas-calc]

Workaround:

spike_train_vector = [(st.t, [unit_id]*len(st)) for unit_id,st in units_tsgroup.items()]
spike_trains = np.concatenate(spike_train_vector, axis=1)
spike_trains.shape # (2, total number of spikes)

#plt.scatter(*spike_trains,s=.1,alpha=.3)
#plt.xlabel("time (s)")
#plt.ylabel("unit id")

times_list,labels_list = spike_trains
labels_list = labels_list.astype(int)
sorting = NumpySorting.from_times_and_labels(times_list=times_list, labels_list=labels_list, sampling_frequency=sampling_frequency)
sorting