def get_features(self, cluster_id, load_all=False):
# Overriden to take into account the sparse structure.
# Only keep spikes belonging to the features spike ids.
if self.features_spike_ids is not None:
# All spikes
spike_ids = self._select_spikes(cluster_id)
spike_ids = np.intersect1d(spike_ids, self.features_spike_ids)
# Relative indices of the spikes in the self.features_spike_ids
# array, necessary to load features from all_features which only
# contains the subset of the spikes.
spike_ids_rel = _index_of(spike_ids, self.features_spike_ids)
else:
spike_ids = self._select_spikes(cluster_id,
self.n_spikes_features
if not load_all else None)
spike_ids_rel = spike_ids
st = self.spike_templates[spike_ids]
nc = self.n_channels
nfpc = self.n_features_per_channel
ns = len(spike_ids)
f = _densify(spike_ids_rel, self.all_features,
self.features_ind[st, :], self.n_channels)
f = np.transpose(f, (0, 2, 1))
assert f.shape == (ns, nc, nfpc)
b = Bunch()
# Normalize features.
m = self.get_feature_lim()
f = _normalize(f, -m, m)
b.data = f
b.spike_ids = spike_ids
b.spike_clusters = self.spike_clusters[spike_ids]
b.masks = self.all_masks[spike_ids]
return b
def get_features(self, cluster_id, load_all=False):
# Overriden to take into account the sparse structure.
# Only keep spikes belonging to the features spike ids.
if self.features_spike_ids is not None:
# All spikes
spike_ids = self._select_spikes(cluster_id)
spike_ids = np.intersect1d(spike_ids, self.features_spike_ids)
# Relative indices of the spikes in the self.features_spike_ids
# array, necessary to load features from all_features which only
# contains the subset of the spikes.
spike_ids_rel = _index_of(spike_ids, self.features_spike_ids)
else:
spike_ids = self._select_spikes(
cluster_id, self.n_spikes_features if not load_all else None)
spike_ids_rel = spike_ids
st = self.spike_templates[spike_ids]
nc = self.n_channels
nfpc = self.n_features_per_channel
ns = len(spike_ids)
f = _densify(spike_ids_rel, self.all_features,
self.features_ind[st, :], self.n_channels)
f = np.transpose(f, (0, 2, 1))
assert f.shape == (ns, nc, nfpc)
b = Bunch()
# Normalize features.
m = self.get_feature_lim()
f = _normalize(f, -m, m)
b.data = f
b.spike_ids = spike_ids
b.spike_clusters = self.spike_clusters[spike_ids]
b.masks = self.all_masks[spike_ids]
return b
def get_features(self, cluster_id, load_all=False):
data = self._select_data(cluster_id,
self.all_features,
(self.n_spikes_features
if not load_all else None),
)
m = self.get_feature_lim()
data.data = _normalize(data.data.copy(), -m, +m)
return data
def get_waveforms(self, cluster_id):
data = self._select_data(cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
)
# Cache the normalized waveforms.
m, M = self.get_waveform_lims()
data.data = _normalize(data.data, m, M)
return [data]
def get_features(self, cluster_id, load_all=False):
data = self._select_data(
cluster_id,
self.all_features,
(self.n_spikes_features if not load_all else None),
)
m = self.get_feature_lim()
data.data = _normalize(data.data.copy(), -m, +m)
return data
def get_waveforms(self, cluster_id):
data = self._select_data(
cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
)
# Cache the normalized waveforms.
m, M = self.get_waveform_lims()
data.data = _normalize(data.data, m, M)
return [data]
def get_waveforms(self, cluster_id):
m, M = self.get_waveform_lims()
if self.all_waveforms is not None:
# Waveforms.
waveforms_b = self._select_data(
cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
)
w = waveforms_b.data
# Sparsify.
channels = np.nonzero(w.mean(axis=1).mean(axis=0))[0]
w = w[:, :, channels]
waveforms_b.channels = channels
# Normalize.
mean = w.mean(axis=1).mean(axis=1)
w = w.astype(np.float64)
w -= mean[:, np.newaxis, np.newaxis]
w = _normalize(w, m, M)
waveforms_b.data = w
waveforms_b.cluster_id = cluster_id
waveforms_b.tag = 'waveforms'
else:
waveforms_b = None
# Find the templates corresponding to the cluster.
template_ids = np.nonzero(self.get_cluster_templates(cluster_id))[0]
# Templates.
templates = self.templates_unw[template_ids]
assert templates.ndim == 3
# Masks.
masks = self.template_masks[template_ids]
assert masks.ndim == 2
assert templates.shape[0] == masks.shape[0]
# Find mean amplitude.
spike_ids = self._select_spikes(cluster_id,
self.n_spikes_waveforms_lim)
mean_amp = self.all_amplitudes[spike_ids].mean()
# Normalize.
# mean = templates.mean(axis=1).mean(axis=1)
templates = templates.astype(np.float64).copy()
# templates -= mean[:, np.newaxis, np.newaxis]
templates *= mean_amp
templates *= 2. / (M - m)
template_b = Bunch(
data=templates,
masks=masks,
alpha=1.,
cluster_id=cluster_id,
tag='templates',
)
if waveforms_b is not None:
return [waveforms_b, template_b]
else:
return [template_b]
def get_background_features(self):
k = max(1, int(self.n_spikes // self.n_spikes_background_features))
spike_ids = slice(None, None, k)
b = Bunch()
b.data = self.all_features[spike_ids]
m = self.get_feature_lim()
b.data = _normalize(b.data.copy(), -m, +m)
b.spike_ids = spike_ids
b.spike_clusters = self.spike_clusters[spike_ids]
b.masks = self.all_masks[spike_ids]
return b
def get_waveforms(self, cluster_id):
m, M = self.get_waveform_lims()
if self.all_waveforms is not None:
# Waveforms.
waveforms_b = self._select_data(
cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
)
mean = waveforms_b.data.mean(axis=1).mean(axis=1)
waveforms_b.data = waveforms_b.data.astype(np.float64)
waveforms_b.data -= mean[:, np.newaxis, np.newaxis]
waveforms_b.data = _normalize(waveforms_b.data, m, M)
else:
waveforms_b = None
# Find the templates corresponding to the cluster.
template_ids = np.nonzero(self.get_cluster_templates(cluster_id))[0]
# Templates.
templates = self.templates_unw[template_ids]
assert templates.ndim == 3
masks = self.template_masks[template_ids]
assert masks.ndim == 2
assert templates.shape[0] == masks.shape[0]
# Find mean amplitude.
spike_ids = self._select_spikes(cluster_id,
self.n_spikes_waveforms_lim)
mean_amp = self.all_amplitudes[spike_ids].mean()
tmp = templates * mean_amp
tmp = _normalize(tmp, m, M)
n = len(template_ids)
template_b = Bunch(
spike_ids=template_ids,
spike_clusters=cluster_id * np.ones(n),
data=tmp,
masks=masks,
alpha=1.,
)
if waveforms_b is not None:
return [waveforms_b, template_b]
else:
return [template_b]
def get_waveforms(self, cluster_id):
m, M = self.get_waveform_lims()
if self.all_waveforms is not None:
# Waveforms.
waveforms_b = self._select_data(cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
)
w = waveforms_b.data
# Sparsify.
channels = np.nonzero(w.mean(axis=1).mean(axis=0))[0]
w = w[:, :, channels]
waveforms_b.channels = channels
# Normalize.
mean = w.mean(axis=1).mean(axis=1)
w = w.astype(np.float64)
w -= mean[:, np.newaxis, np.newaxis]
w = _normalize(w, m, M)
waveforms_b.data = w
waveforms_b.cluster_id = cluster_id
waveforms_b.tag = 'waveforms'
else:
waveforms_b = None
# Find the templates corresponding to the cluster.
template_ids = np.nonzero(self.get_cluster_templates(cluster_id))[0]
# Templates.
templates = self.templates_unw[template_ids]
assert templates.ndim == 3
# Masks.
masks = self.template_masks[template_ids]
assert masks.ndim == 2
assert templates.shape[0] == masks.shape[0]
# Find mean amplitude.
spike_ids = self._select_spikes(cluster_id,
self.n_spikes_waveforms_lim)
mean_amp = self.all_amplitudes[spike_ids].mean()
# Normalize.
# mean = templates.mean(axis=1).mean(axis=1)
templates = templates.astype(np.float64).copy()
# templates -= mean[:, np.newaxis, np.newaxis]
templates *= mean_amp
templates *= 2. / (M - m)
template_b = Bunch(data=templates,
masks=masks,
alpha=1.,
cluster_id=cluster_id,
tag='templates',
)
if waveforms_b is not None:
return [waveforms_b, template_b]
else:
return [template_b]
def get_waveforms(self, cluster_id):
m, M = self.get_waveform_lims()
if self.all_waveforms is not None:
# Waveforms.
waveforms_b = self._select_data(cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
)
mean = waveforms_b.data.mean(axis=1).mean(axis=1)
waveforms_b.data = waveforms_b.data.astype(np.float64)
waveforms_b.data -= mean[:, np.newaxis, np.newaxis]
waveforms_b.data = _normalize(waveforms_b.data, m, M)
else:
waveforms_b = None
# Find the templates corresponding to the cluster.
template_ids = np.nonzero(self.get_cluster_templates(cluster_id))[0]
# Templates.
templates = self.templates_unw[template_ids]
assert templates.ndim == 3
masks = self.template_masks[template_ids]
assert masks.ndim == 2
assert templates.shape[0] == masks.shape[0]
# Find mean amplitude.
spike_ids = self._select_spikes(cluster_id,
self.n_spikes_waveforms_lim)
mean_amp = self.all_amplitudes[spike_ids].mean()
tmp = templates * mean_amp
tmp = _normalize(tmp, m, M)
n = len(template_ids)
template_b = Bunch(spike_ids=template_ids,
spike_clusters=cluster_id * np.ones(n),
data=tmp,
masks=masks,
alpha=1.,
)
if waveforms_b is not None:
return [waveforms_b, template_b]
else:
return [template_b]
def get_waveforms(self, cluster_id):
data = self._select_data(cluster_id,
self.all_waveforms,
self.n_spikes_waveforms,
batch_size=10,
)
# Sparsify the waveforms.
w = data.data
channels = np.nonzero(w.mean(axis=1).mean(axis=0))[0]
# FIX when the waveforms are null because of absent raw data.
if not len(channels): # pragma: no cover
channels = np.nonzero(self.get_mean_masks(cluster_id) > 0)[0]
w = w[:, :, channels]
data.channels = channels
# Cache the normalized waveforms.
m, M = self.get_waveform_lims()
data.data = _normalize(w, m, M)
data.cluster_id = cluster_id
return data
