def test_cluster_selection():
indices = [10, 20, 25]
clusters = generate_clusters(indices)
clusters_selected = [1, 2]
spikes = get_spikes_in_clusters(clusters_selected, clusters, False)
assert np.array_equal(np.nonzero(spikes)[0], indices)
spikes = get_spikes_in_clusters(clusters_selected, clusters, True)
assert np.array_equal(spikes, indices)
def test_select_array():
# All spikes in cluster 1.
indices = [10, 20, 25]
# Indices in data excerpt.
indices_data = [5, 10, 15, 20]
# Generate clusters and data.
clusters = generate_clusters(indices)
data_raw = generate_data2D()
data = pd.DataFrame(data_raw)
# Excerpt of the data.
data_excerpt = select(data, indices_data)
# Get all spike indices in cluster 1.
spikes_inclu1 = get_spikes_in_clusters([1], clusters)
# We want to select all clusters in cluster 1 among those in data excerpt.
data_excerpt_inclu1 = select(data_excerpt, spikes_inclu1)
# There should be two rows: 4 in the excerpt, among which two are in
# cluster 1.
assert data_excerpt_inclu1.shape == (2, 5)
def cluster_quality(waveforms, features, clusters, masks,
clusters_selected=None):
# clusters = select(clusters, spikes)
# features = select(features, spikes)
# waveforms = select(waveforms, spikes)
# masks = select(masks, spikes)
nspikes, nsamples, nchannels = waveforms.shape
quality = {}
for cluster in clusters_selected:
spikes = get_spikes_in_clusters(cluster, clusters)
w = select(waveforms, spikes)
m = select(masks, spikes)
q = 1. / nsamples * ((w ** 2).sum(axis=1) * 1).mean(axis=1).max()
quality[cluster] = q
return quality
def cluster_quality(waveforms, features, clusters, masks,
clusters_selected=None):
# clusters = select(clusters, spikes)
# features = select(features, spikes)
# waveforms = select(waveforms, spikes)
# masks = select(masks, spikes)
nspikes, nsamples, nchannels = waveforms.shape
quality = {}
for cluster in clusters_selected:
spikes = get_spikes_in_clusters(cluster, clusters)
w = select(waveforms, spikes)
m = select(masks, spikes)
q = 1. / nsamples * ((w ** 2).sum(axis=1) * 1).mean(axis=1).max()
quality[cluster] = q
return quality
def prepare_average_waveform_data(self):
waveforms_avg = np.zeros((self.nclusters, self.nsamples, self.nchannels))
waveforms_std = np.zeros((self.nclusters, self.nsamples, self.nchannels))
self.masks_avg = np.zeros((self.nclusters, self.nchannels))
for i, cluster in enumerate(self.clusters_unique):
spike_indices = get_spikes_in_clusters(cluster, self.clusters)
w = select(self.waveforms, spike_indices)
m = select(self.masks, spike_indices)
waveforms_avg[i,...] = w.mean(axis=0)
waveforms_std[i,...] = w.std(axis=0).mean()
self.masks_avg[i,...] = m.mean(axis=0)
# create X coordinates
X = np.tile(np.linspace(-1., 1., self.nsamples),
(self.nchannels * self.nclusters, 1))
# create Y coordinates
if self.nclusters == 0:
Y = np.array([], dtype=np.float32)
thickness = np.array([], dtype=np.float32)
else:
Y = np.vstack(waveforms_avg)
thickness = np.vstack(waveforms_std).T.ravel()
# concatenate data
data = np.empty((X.size, 2), dtype=np.float32)
data[:,0] = X.ravel()
data[:,1] = Y.T.ravel()
if self.nclusters > 0:
# thicken
w = thickness.reshape((-1, 1))
n = waveforms_avg.size
Y = np.zeros((2 * n, 2))
u = np.zeros((n, 2))
u[1:,0] = -np.diff(data[:,1])
u[1:,1] = data[1,0] - data[0,0]
u[0,:] = u[1,:]
r = (u[:,0] ** 2 + u[:,1] ** 2) ** .5
r[r == 0.] = 1
u[:,0] /= r
u[:,1] /= r
Y[::2,:] = data - w * u
Y[1::2,:] = data + w * u
data_thickened = Y
else:
n = 0
data_thickened = data
self.nsamples_avg = self.nsamples * 2
self.npoints_avg = waveforms_avg.size * 2
self.nspikes_avg = self.nclusters
self.nclusters_avg = self.nclusters
self.nchannels_avg = self.nchannels
self.clusters_rel_avg = np.arange(self.nclusters, dtype=np.int32)
self.clusters_rel_ordered_avg = np.argsort(self.clusters_selected)[self.clusters_rel_avg]
if self.keep_order:
self.clusters_rel_ordered_avg2 = self.clusters_rel_ordered_avg
else:
self.clusters_rel_ordered_avg2 = self.clusters_rel_avg
return data_thickened
def prepare_average_waveform_data(self):
waveforms_avg = np.zeros((self.nclusters, self.nsamples, self.nchannels))
waveforms_std = np.zeros((self.nclusters, self.nsamples, self.nchannels))
self.masks_avg = np.zeros((self.nclusters, self.nchannels))
for i, cluster in enumerate(self.clusters_unique):
spike_indices = get_spikes_in_clusters(cluster, self.clusters)
w = select(self.waveforms, spike_indices)
m = select(self.masks, spike_indices)
waveforms_avg[i,...] = w.mean(axis=0)
waveforms_std[i,...] = w.std(axis=0).mean()
self.masks_avg[i,...] = m.mean(axis=0)
# create X coordinates
X = np.tile(np.linspace(-1., 1., self.nsamples),
(self.nchannels * self.nclusters, 1))
# create Y coordinates
if self.nclusters == 0:
Y = np.array([], dtype=np.float32)
thickness = np.array([], dtype=np.float32)
else:
Y = np.vstack(waveforms_avg)
thickness = np.vstack(waveforms_std).T.ravel()
# concatenate data
data = np.empty((X.size, 2), dtype=np.float32)
data[:,0] = X.ravel()
data[:,1] = Y.T.ravel()
if self.nclusters > 0:
# thicken
w = thickness.reshape((-1, 1))
n = waveforms_avg.size
Y = np.zeros((2 * n, 2))
u = np.zeros((n, 2))
u[1:,0] = -np.diff(data[:,1])
u[1:,1] = data[1,0] - data[0,0]
u[0,:] = u[1,:]
r = (u[:,0] ** 2 + u[:,1] ** 2) ** .5
r[r == 0.] = 1
u[:,0] /= r
u[:,1] /= r
Y[::2,:] = data - w * u
Y[1::2,:] = data + w * u
data_thickened = Y
else:
n = 0
data_thickened = data
self.nsamples_avg = self.nsamples * 2
self.npoints_avg = waveforms_avg.size * 2
self.nspikes_avg = self.nclusters
self.nclusters_avg = self.nclusters
self.nchannels_avg = self.nchannels
self.clusters_rel_avg = np.arange(self.nclusters, dtype=np.int32)
self.clusters_rel_ordered_avg = np.argsort(self.clusters_selected)[self.clusters_rel_avg]
if self.keep_order:
self.clusters_rel_ordered_avg2 = self.clusters_rel_ordered_avg
else:
self.clusters_rel_ordered_avg2 = self.clusters_rel_avg
return data_thickened