Note
Click here to download the full example code
SortingExtractor objects¶
The SortingExtractor
is the basic class for handling spike sorted data.
Here is how it works.
A SortingExtractor handles:
spike trains retrieval across segments
dumping to/loading from dict-json
saving (caching)
import numpy as np
import spikeinterface.extractors as se
We will create a SortingExtractor
object from scratch using numpy
and the
NumpySorting
Let’s define the properties of the dataset:
sampling_frequency = 30000.
duration = 20.
num_timepoints = int(sampling_frequency * duration)
num_units = 4
num_spikes = 1000
We generate some random events for 2 segments:
times0 = np.int_(np.sort(np.random.uniform(0, num_timepoints, num_spikes)))
labels0 = np.random.randint(1, num_units + 1, size=num_spikes)
times1 = np.int_(np.sort(np.random.uniform(0, num_timepoints, num_spikes)))
labels1 = np.random.randint(1, num_units + 1, size=num_spikes)
And instantiate a NumpySorting
object:
sorting = se.NumpySorting.from_times_labels([times0, times1], [labels0, labels1], sampling_frequency)
print(sorting)
Out:
NumpySorting: 4 units - 2 segments - 30.0kHz
We can now print properties that the SortingExtractor
retrieves from
the underlying sorted dataset.
print('Unit ids = {}'.format(sorting.get_unit_ids()))
st = sorting.get_unit_spike_train(unit_id=1, segment_index=0)
print('Num. events for unit 1seg0 = {}'.format(len(st)))
st1 = sorting.get_unit_spike_train(unit_id=1, start_frame=0, end_frame=30000, segment_index=1)
print('Num. events for first second of unit 1 seg1 = {}'.format(len(st1)))
Out:
Unit ids = [1 2 3 4]
Num. events for unit 1seg0 = 252
Num. events for first second of unit 1 seg1 = 5
Some extractors also implement a write
function. We can for example
save our newly created sorting object to NPZ format (a simple format based
on numpy used in spikeinterface
):
file_path = 'my_sorting.npz'
se.NpzSortingExtractor.write_sorting(sorting, file_path)
We can now read it back with the proper extractor:
sorting2 = se.NpzSortingExtractor(file_path)
print(sorting2)
Out:
NpzSortingExtractor: 4 units - 2 segments - 30.0kHz
file_path: /home/docs/checkouts/readthedocs.org/user_builds/spikeinterface/checkouts/0.90.1/examples/modules/core/my_sorting.npz
Unit properties are key value pairs that we can store for any unit.
We will now calculate unit firing rates and add them as properties to
the SortingExtractor
object:
firing_rates = []
for unit_id in sorting2.get_unit_ids():
st = sorting2.get_unit_spike_train(unit_id=unit_id, segment_index=0)
firing_rates.append(st.size / duration)
sorting2.set_property('firing_rate', firing_rates)
print(sorting2.get_property('firing_rate'))
Out:
[12.6 12.5 12.25 12.65]
You can also get a a sorting with a subset of unit. Properties are propagated to the new object:
sorting3 = sorting2.select_units(unit_ids=[1, 4])
print(sorting3)
print(sorting3.get_property('firing_rate'))
# which is equivalent to
from spikeinterface import UnitsSelectionSorting
sorting3 = UnitsSelectionSorting(sorting2, unit_ids=[1, 4])
Out:
UnitsSelectionSorting: 2 units - 2 segments - 30.0kHz
[12.6 12.65]
- A sorting can be “dumped” (exported) to:
a dict
- * a json file
a pickle file
The “dump” operation is lazy, i.e., the spike trains are not exported. Only the information about how to reconstruct the sorting are dumped:
from spikeinterface import load_extractor
from pprint import pprint
d = sorting2.to_dict()
pprint(d)
sorting2_loaded = load_extractor(d)
print(sorting2_loaded)
Out:
{'annotations': {},
'class': 'spikeinterface.core.npzsortingextractor.NpzSortingExtractor',
'dumpable': True,
'kwargs': {'file_path': '/home/docs/checkouts/readthedocs.org/user_builds/spikeinterface/checkouts/0.90.1/examples/modules/core/my_sorting.npz'},
'module': 'spikeinterface',
'properties': {},
'relative_paths': False,
'version': '0.90.1'}
NpzSortingExtractor: 4 units - 2 segments - 30.0kHz
file_path: /home/docs/checkouts/readthedocs.org/user_builds/spikeinterface/checkouts/0.90.1/examples/modules/core/my_sorting.npz
The dictionary can also be dumped directly to a JSON file on disk:
sorting2.dump('my_sorting.json')
sorting2_loaded = load_extractor('my_sorting.json')
print(sorting2_loaded)
Out:
NpzSortingExtractor: 4 units - 2 segments - 30.0kHz
file_path: /home/docs/checkouts/readthedocs.org/user_builds/spikeinterface/checkouts/0.90.1/examples/modules/core/my_sorting.npz
IMPORTANT: the “dump” operation DOES NOT copy the spike trains to disk!
If you wish to also store the spike trains in a compact way you need to use the
save()
function:
sorting2.save(folder='./my_sorting')
import os
pprint(os.listdir('./my_sorting'))
sorting2_cached = load_extractor('./my_sorting')
print(sorting2_cached)
Out:
['sorting_cached.npz', 'properties', 'cached.json', 'provenance.json']
NpzSortingExtractor: 4 units - 2 segments - 30.0kHz
file_path: /home/docs/checkouts/readthedocs.org/user_builds/spikeinterface/checkouts/0.90.1/examples/modules/core/my_sorting/sorting_cached.npz
Total running time of the script: ( 0 minutes 0.022 seconds)