def test_filter_unsigned():
traces = np.random.randint(1, 1000, (5000, 4), dtype="uint16")
rec = NumpyRecording(traces_list=traces, sampling_frequency=1000)
rec = rec.save()
rec2 = bandpass_filter(rec, freq_min=10., freq_max=300.)
assert not np.issubdtype(rec2.get_dtype(), np.unsignedinteger)
traces2 = rec2.get_traces()
assert not np.issubdtype(traces2.dtype, np.unsignedinteger)
# notch filter note supported for unsigned
with pytest.raises(TypeError):
rec3 = notch_filter(rec, freq=300., q=10)
# this is ok
rec3 = notch_filter(rec, freq=300., q=10, dtype="float32")
def test_remove_bad_channels():
num_channels = 4
sampling_frequency = 30000.
durations = [10.325, 3.5]
num_segments = len(durations)
num_timepoints = [int(sampling_frequency * d) for d in durations]
traces_list = []
for i in range(num_segments):
traces = np.random.randn(num_timepoints[i], num_channels).astype('float32')
# one channel have big noise
traces[:, 1] *= 10
times = np.arange(num_timepoints[i]) / sampling_frequency
traces += np.sin(2 * np.pi * 50 * times)[:, None]
traces_list.append(traces)
rec = NumpyRecording(traces_list, sampling_frequency)
probe = generate_linear_probe(num_elec=num_channels)
probe.set_device_channel_indices(np.arange(num_channels))
rec.set_probe(probe, in_place=True)
rec2 = remove_bad_channels(rec, bad_threshold=5.)
# Check that the noisy channel is taken out
assert np.array_equal(rec2.get_channel_ids(), [0, 2, 3]), "wrong channel detected."
# Check that the number of segments is maintained after preprocessor
assert np.array_equal(rec2.get_num_segments(), rec.get_num_segments()), "wrong numbber of segments."
# Check that the size of the segments os maintained after preprocessor
assert np.array_equal(*([r.get_num_frames(x) for x in range(rec.get_num_segments())] for r in
[rec, rec2])), "wrong lenght of resulting segments."
# Check that locations are mantained
assert np.array_equal(rec.get_channel_locations()[[0, 2, 3]],
rec2.get_channel_locations()), "wrong channels locations."
def generate_recording(
num_channels=2,
sampling_frequency=30000., # in Hz
durations=[10.325, 3.5], # in s for 2 segments
):
num_segments = len(durations)
num_timepoints = [int(sampling_frequency * d) for d in durations]
traces_list = []
for i in range(num_segments):
traces = np.random.randn(num_timepoints[i],
num_channels).astype('float32')
times = np.arange(num_timepoints[i]) / sampling_frequency
traces += np.sin(2 * np.pi * 50 * times)[:, None]
traces_list.append(traces)
recording = NumpyRecording(traces_list, sampling_frequency)
probe = generate_linear_probe(num_elec=num_channels)
probe.set_device_channel_indices(np.arange(num_channels))
recording.set_probe(probe, in_place=True)
return recording
import matplotlib.pyplot as plt import numpy as np from spikeinterface import NumpyRecording, NumpySorting from spikeinterface import append_recordings, concatenate_recordings ############################################################################## # First let's generate 2 recordings with 2 and 3 segments respectively. # sampling_frequency = 1000. trace0 = np.zeros((150, 5), dtype='float32') trace1 = np.zeros((100, 5), dtype='float32') rec0 = NumpyRecording([trace0, trace1], sampling_frequency) print(rec0) trace2 = np.zeros((50, 5), dtype='float32') trace3 = np.zeros((200, 5), dtype='float32') trace4 = np.zeros((120, 5), dtype='float32') rec1 = NumpyRecording([trace2, trace3, trace4], sampling_frequency) print(rec1) ############################################################################## # lets use `append_recordings()` recording_list = [rec0, rec1] rec = append_recordings(recording_list) print(rec) for i in range(rec.get_num_segments()):