Sorting analyzer does not propagate sparsity to template extension

[h-mayorquin]

from spikeinterface.core import create_sorting_analyzer

from spikeinterface.core import generate_ground_truth_recording

job_kwargs = dict(
    n_jobs=1,
    progress_bar=False,
    verbose=False,
    chunk_duration=1.0,
)

recording, sorting = generate_ground_truth_recording(num_channels=384, durations=[10], seed=0)

analyzer = create_sorting_analyzer(sorting, recording, sparse=True, format="memory", **job_kwargs)

random_spike_parameters = {
    "method": "all",
}

template_extension_parameters = {
    "ms_before": 3.0,
    "ms_after": 5.0,
    "operators": ["average"],
}

extensions = {
    "random_spikes": random_spike_parameters,
    "templates": template_extension_parameters,
}

analyzer.compute_several_extensions(
    extensions=extensions,
    **job_kwargs
)

template = analyzer.get_extension("templates").get_data(outputs="Templates")
template.sparsity
template.are_templates_sparse()

Outputs False

[h-mayorquin]

It seems to be the case also even if computing through the waveforms extension:

from spikeinterface.core import create_sorting_analyzer

from spikeinterface.core import generate_ground_truth_recording

job_kwargs = dict(
    n_jobs=1,
    progress_bar=True,
    verbose=True,
    chunk_duration=1.0,
)

recording, sorting = generate_ground_truth_recording(num_channels=384, durations=[10], seed=0)

analyzer = create_sorting_analyzer(sorting, recording, sparse=True, format="memory", **job_kwargs)

random_spike_parameters = {
    "method": "all",
}

waveform_extension_parameters = {
    "ms_before": 3.0,
    "ms_after": 5.0,
}

template_extension_parameters = {
    "ms_before": 3.0,
    "ms_after": 5.0,
    "operators": ["average"],
}

extensions = {
    "random_spikes": random_spike_parameters,
    "waveforms": waveform_extension_parameters,
    "templates": template_extension_parameters,
}

analyzer.compute_several_extensions(
    extensions=extensions,
    **job_kwargs
)

print(analyzer.get_extension("waveforms").get_data().shape)
templates = analyzer.get_extension("templates").get_data(outputs="Templates")
templates.are_templates_sparse()

Output:

(1471, 200, 20)  # So waveforms are sparse
False

[samuelgarcia]

Yes I knew. This is because estimate_templates_with_accumulator() is not sparse. At the moment template are represented dense always internally, even with sparse we put zeros. We should change this but this was to make it compatible with the waveforms extractor.