Allow for concatenated mode in get_some_projections and compute_pc_metrics

[chrishalcrow]

Should fix #3653

Fixes bug where get_some_projections and compute_pc_metrics couldn't extract pcas when they were computed in concatenated mode. Fixing this allows pca quality metrics to be computed too.

Added a test which used to fail, and now passes.

Harder to test if this definitely works. Here's some evidence: it's code which computes the quality metrics in concatenated mode (with 20 components) and in non-concatenated mode (with 5x4=20 components), then computes the metrics. We wouldn't expect the results to be equal, but would expect them to be highly correlated, which they seem to be!

import numpy as np
import spikeinterface.full as si

rec, sort = si.generate_ground_truth_recording()

sa_1 = si.create_sorting_analyzer(recording=rec, sorting=sort, folder="my_sa", format="binary_folder", overwrite=True, sparse=False, seed=1205)

sa_1.compute({
    "noise_levels": {},
    "random_spikes": {'seed': 1205},
    "waveforms": {},
    "templates": {},
    "principal_components": {"mode": 'concatenated', "n_components": 20},
    "quality_metrics": {"metric_names": ["d_prime","silhouette", "nearest_neighbor"]}
})

sa_2 = si.create_sorting_analyzer(recording=rec, sorting=sort, folder="my_sa", format="binary_folder", overwrite=True, sparse=False)

sa_2.compute({
    "noise_levels": {},
    "random_spikes": {'seed': 1205},
    "waveforms": {},
    "templates": {},
    "principal_components": {},
    "quality_metrics": {"metric_names": ["d_prime","silhouette", "nearest_neighbor"]}
})

qm_1 = sa_1.get_extension("quality_metrics").get_data()
qm_2 = sa_2.get_extension("quality_metrics").get_data()

dprime_corr = np.corrcoef(x = [ qm_1['d_prime'].values, qm_2['d_prime'].values ])
sil_corr = np.corrcoef(x = [ qm_1['silhouette'].values, qm_2['silhouette'].values ])
nnhit_corr = np.corrcoef(x = [ qm_1['nn_hit_rate'].values, qm_2['nn_hit_rate'].values ])
nnmiss_corr = np.corrcoef(x = [ qm_1['nn_miss_rate'].values, qm_2['nn_miss_rate'].values ])

print(dprime_corr)
print(sil_corr)
print(nnhit_corr)
print(nnmiss_corr)

with output:

dprime_corr:  [[1.         0.99704657]
 [0.99704657 1.        ]]
sil_corr:  [[1.         0.95037167]
 [0.95037167 1.        ]]
nnhit_corr:  [[1.         0.98622277]
 [0.98622277 1.        ]]
nnmiss_corr:  [[1.         0.96420438]
 [0.96420438 1.        ]]

[b-grimaud]

Just tested it and it works, thanks !