def _load_traces(self, channel_map=None):
traces = load_raw_data(self.dat_path,
n_channels_dat=self.n_channels_dat,
dtype=self.dtype,
offset=self.offset,
)
if traces is not None:
# Find the scaling factor for the traces.
traces = _concatenate_virtual_arrays([traces],
channel_map,
)
return traces
def _init_data(self):
if op.exists(self.dat_path):
logger.debug("Loading traces at `%s`.", self.dat_path)
traces = _dat_to_traces(self.dat_path,
n_channels=self.n_channels_dat,
dtype=self.dtype or np.int16,
offset=self.offset,
)
n_samples_t, _ = traces.shape
assert _ == self.n_channels_dat
else:
traces = None
n_samples_t = 0
logger.debug("Loading amplitudes.")
amplitudes = read_array('amplitudes').squeeze()
n_spikes, = amplitudes.shape
self.n_spikes = n_spikes
# Create spike_clusters if the file doesn't exist.
if not op.exists(filenames['spike_clusters']):
shutil.copy(filenames['spike_templates'],
filenames['spike_clusters'])
logger.debug("Loading %d spike clusters.", self.n_spikes)
spike_clusters = read_array('spike_clusters').squeeze()
spike_clusters = spike_clusters.astype(np.int32)
assert spike_clusters.shape == (n_spikes,)
self.spike_clusters = spike_clusters
logger.debug("Loading spike templates.")
spike_templates = read_array('spike_templates').squeeze()
spike_templates = spike_templates.astype(np.int32)
assert spike_templates.shape == (n_spikes,)
self.spike_templates = spike_templates
logger.debug("Loading spike samples.")
spike_samples = read_array('spike_samples').squeeze()
assert spike_samples.shape == (n_spikes,)
logger.debug("Loading templates.")
templates = read_array('templates')
templates[np.isnan(templates)] = 0
# templates = np.transpose(templates, (2, 1, 0))
# Unwhiten the templates.
logger.debug("Loading the whitening matrix.")
self.whitening_matrix = read_array('whitening_matrix')
if op.exists(filenames['templates_unw']):
logger.debug("Loading unwhitened templates.")
templates_unw = read_array('templates_unw')
templates_unw[np.isnan(templates_unw)] = 0
else:
logger.debug("Couldn't find unwhitened templates, computing them.")
logger.debug("Inversing the whitening matrix %s.",
self.whitening_matrix.shape)
wmi = np.linalg.inv(self.whitening_matrix)
logger.debug("Unwhitening the templates %s.",
templates.shape)
templates_unw = np.dot(np.ascontiguousarray(templates),
np.ascontiguousarray(wmi))
# Save the unwhitened templates.
write_array('templates_unw.npy', templates_unw)
n_templates, n_samples_templates, n_channels = templates.shape
self.n_templates = n_templates
logger.debug("Loading similar templates.")
self.similar_templates = read_array('similar_templates')
assert self.similar_templates.shape == (self.n_templates,
self.n_templates)
logger.debug("Loading channel mapping.")
channel_mapping = read_array('channel_mapping').squeeze()
channel_mapping = channel_mapping.astype(np.int32)
assert channel_mapping.shape == (n_channels,)
# Ensure that the mappings maps to valid columns in the dat file.
assert np.all(channel_mapping <= self.n_channels_dat - 1)
logger.debug("Loading channel positions.")
channel_positions = read_array('channel_positions')
assert channel_positions.shape == (n_channels, 2)
if op.exists(filenames['features']):
logger.debug("Loading features.")
all_features = np.load(filenames['features'], mmap_mode='r')
features_ind = read_array('features_ind').astype(np.int32)
# Feature subset.
if op.exists(filenames['features_spike_ids']):
features_spike_ids = read_array('features_spike_ids') \
.astype(np.int32)
assert len(features_spike_ids) == len(all_features)
self.features_spike_ids = features_spike_ids
ns = len(features_spike_ids)
else:
ns = self.n_spikes
self.features_spike_ids = None
assert all_features.ndim == 3
n_loc_chan = all_features.shape[2]
self.n_features_per_channel = all_features.shape[1]
assert all_features.shape == (ns,
self.n_features_per_channel,
n_loc_chan,
)
# Check sparse features arrays shapes.
assert features_ind.shape == (self.n_templates, n_loc_chan)
else:
all_features = None
features_ind = None
self.all_features = all_features
self.features_ind = features_ind
if op.exists(filenames['template_features']):
logger.debug("Loading template features.")
template_features = np.load(filenames['template_features'],
mmap_mode='r')
template_features_ind = read_array('template_features_ind'). \
astype(np.int32)
template_features_ind = template_features_ind.copy()
n_sim_tem = template_features.shape[1]
assert template_features.shape == (n_spikes, n_sim_tem)
assert template_features_ind.shape == (n_templates, n_sim_tem)
else:
template_features = None
template_features_ind = None
self.template_features_ind = template_features_ind
self.template_features = template_features
self.n_channels = n_channels
# Take dead channels into account.
if traces is not None:
# Find the scaling factor for the traces.
scaling = 1. / self._data_lim(traces[:10000])
traces = _concatenate_virtual_arrays([traces],
channel_mapping,
scaling=scaling,
)
else:
scaling = 1.
# Amplitudes
self.all_amplitudes = amplitudes
self.amplitudes_lim = self.all_amplitudes.max()
# Templates
self.templates = templates
self.templates_unw = templates_unw
assert self.templates.shape == self.templates_unw.shape
self.n_samples_templates = n_samples_templates
self.n_samples_waveforms = n_samples_templates
self.template_lim = np.max(np.abs(self.templates))
self.duration = n_samples_t / float(self.sample_rate)
self.spike_times = spike_samples / float(self.sample_rate)
assert np.all(np.diff(self.spike_times) >= 0)
self.cluster_ids = np.unique(self.spike_clusters)
# n_clusters = len(self.cluster_ids)
self.channel_positions = channel_positions
self.all_traces = traces
# Filter the waveforms.
order = 3
filter_margin = order * 3
b_filter = bandpass_filter(rate=self.sample_rate,
low=500.,
high=self.sample_rate * .475,
order=order)
# Only filter the data for the waveforms if the traces
# are not already filtered.
if not getattr(self, 'hp_filtered', False):
logger.debug("HP filtering the data for waveforms")
def the_filter(x, axis=0):
return apply_filter(x, b_filter, axis=axis)
else:
the_filter = None
# Fetch waveforms from traces.
nsw = self.n_samples_waveforms
if traces is not None:
waveforms = WaveformLoader(traces=traces,
n_samples_waveforms=nsw,
filter=the_filter,
filter_margin=filter_margin,
)
waveforms = SpikeLoader(waveforms, spike_samples)
else:
waveforms = None
self.all_waveforms = waveforms
self.template_masks = get_masks(self.templates)
self.all_masks = MaskLoader(self.template_masks, self.spike_templates)
# Read the cluster groups.
logger.debug("Loading the cluster groups.")
self.cluster_groups = {}
if op.exists(filenames['cluster_groups']):
with open(filenames['cluster_groups'], 'r') as f:
reader = csv.reader(f, delimiter='\t')
# Skip the header.
for row in reader:
break
for row in reader:
cluster, group = row
cluster = int(cluster)
self.cluster_groups[cluster] = group
for cluster_id in self.cluster_ids:
if cluster_id not in self.cluster_groups:
self.cluster_groups[cluster_id] = None
def _init_data(self):
if op.exists(self.dat_path):
logger.debug("Loading traces at `%s`.", self.dat_path)
traces = _dat_to_traces(
self.dat_path,
n_channels=self.n_channels_dat,
dtype=self.dtype or np.int16,
offset=self.offset,
)
n_samples_t, _ = traces.shape
assert _ == self.n_channels_dat
else:
if self.dat_path is not None:
logger.warning("Error while loading data: File %s not found.",
self.dat_path)
traces = None
n_samples_t = 0
logger.debug("Loading amplitudes.")
amplitudes = read_array('amplitudes').squeeze()
n_spikes, = amplitudes.shape
self.n_spikes = n_spikes
# Create spike_clusters if the file doesn't exist.
if not op.exists(filenames['spike_clusters']):
shutil.copy(filenames['spike_templates'],
filenames['spike_clusters'])
logger.debug("Loading %d spike clusters.", self.n_spikes)
spike_clusters = read_array('spike_clusters').squeeze()
spike_clusters = spike_clusters.astype(np.int32)
assert spike_clusters.shape == (n_spikes, )
self.spike_clusters = spike_clusters
logger.debug("Loading spike templates.")
spike_templates = read_array('spike_templates').squeeze()
spike_templates = spike_templates.astype(np.int32)
assert spike_templates.shape == (n_spikes, )
self.spike_templates = spike_templates
logger.debug("Loading spike samples.")
spike_samples = read_array('spike_samples').squeeze()
assert spike_samples.shape == (n_spikes, )
logger.debug("Loading templates.")
templates = read_array('templates')
templates[np.isnan(templates)] = 0
# templates = np.transpose(templates, (2, 1, 0))
# Unwhiten the templates.
logger.debug("Loading the whitening matrix.")
self.whitening_matrix = read_array('whitening_matrix')
if op.exists(filenames['templates_unw']):
logger.debug("Loading unwhitened templates.")
templates_unw = read_array('templates_unw')
templates_unw[np.isnan(templates_unw)] = 0
else:
logger.debug("Couldn't find unwhitened templates, computing them.")
logger.debug("Inversing the whitening matrix %s.",
self.whitening_matrix.shape)
wmi = np.linalg.inv(self.whitening_matrix)
logger.debug("Unwhitening the templates %s.", templates.shape)
templates_unw = np.dot(np.ascontiguousarray(templates),
np.ascontiguousarray(wmi))
# Save the unwhitened templates.
write_array('templates_unw.npy', templates_unw)
n_templates, n_samples_templates, n_channels = templates.shape
self.n_templates = n_templates
logger.debug("Loading similar templates.")
self.similar_templates = read_array('similar_templates')
assert self.similar_templates.shape == (self.n_templates,
self.n_templates)
logger.debug("Loading channel mapping.")
channel_mapping = read_array('channel_mapping').squeeze()
channel_mapping = channel_mapping.astype(np.int32)
assert channel_mapping.shape == (n_channels, )
# Ensure that the mappings maps to valid columns in the dat file.
assert np.all(channel_mapping <= self.n_channels_dat - 1)
logger.debug("Loading channel positions.")
channel_positions = read_array('channel_positions')
assert channel_positions.shape == (n_channels, 2)
if op.exists(filenames['features']):
logger.debug("Loading features.")
all_features = np.load(filenames['features'], mmap_mode='r')
features_ind = read_array('features_ind').astype(np.int32)
# Feature subset.
if op.exists(filenames['features_spike_ids']):
features_spike_ids = read_array('features_spike_ids') \
.astype(np.int32)
assert len(features_spike_ids) == len(all_features)
self.features_spike_ids = features_spike_ids
ns = len(features_spike_ids)
else:
ns = self.n_spikes
self.features_spike_ids = None
assert all_features.ndim == 3
n_loc_chan = all_features.shape[2]
self.n_features_per_channel = all_features.shape[1]
assert all_features.shape == (
ns,
self.n_features_per_channel,
n_loc_chan,
)
# Check sparse features arrays shapes.
assert features_ind.shape == (self.n_templates, n_loc_chan)
else:
all_features = None
features_ind = None
self.all_features = all_features
self.features_ind = features_ind
if op.exists(filenames['template_features']):
logger.debug("Loading template features.")
template_features = np.load(filenames['template_features'],
mmap_mode='r')
template_features_ind = read_array('template_features_ind'). \
astype(np.int32)
template_features_ind = template_features_ind.copy()
n_sim_tem = template_features.shape[1]
assert template_features.shape == (n_spikes, n_sim_tem)
assert template_features_ind.shape == (n_templates, n_sim_tem)
else:
template_features = None
template_features_ind = None
self.template_features_ind = template_features_ind
self.template_features = template_features
self.n_channels = n_channels
# Take dead channels into account.
if traces is not None:
# Find the scaling factor for the traces.
scaling = 1. / self._data_lim(traces[:10000])
traces = _concatenate_virtual_arrays(
[traces],
channel_mapping,
scaling=scaling,
)
else:
scaling = 1.
# Amplitudes
self.all_amplitudes = amplitudes
self.amplitudes_lim = np.max(self.all_amplitudes)
# Templates
self.templates = templates
self.templates_unw = templates_unw
assert self.templates.shape == self.templates_unw.shape
self.n_samples_templates = n_samples_templates
self.n_samples_waveforms = n_samples_templates
self.template_lim = np.max(np.abs(self.templates))
self.duration = n_samples_t / float(self.sample_rate)
self.spike_times = spike_samples / float(self.sample_rate)
assert np.all(np.diff(self.spike_times) >= 0)
self.cluster_ids = _unique(self.spike_clusters)
# n_clusters = len(self.cluster_ids)
self.channel_positions = channel_positions
self.all_traces = traces
# Only filter the data for the waveforms if the traces
# are not already filtered.
if not getattr(self, 'hp_filtered', False):
logger.debug("HP filtering the data for waveforms")
filter_order = 3
else:
filter_order = None
n_closest_channels = getattr(self, 'max_n_unmasked_channels', 16)
mask_threshold = getattr(self, 'waveform_mask_threshold', None)
self.closest_channels = get_closest_channels(
self.channel_positions,
n_closest_channels,
)
self.template_masks = get_masks(self.templates, self.closest_channels)
self.all_masks = MaskLoader(self.template_masks, self.spike_templates)
# Fetch waveforms from traces.
nsw = self.n_samples_waveforms
if traces is not None:
waveforms = WaveformLoader(
traces=traces,
masks=self.all_masks,
spike_samples=spike_samples,
n_samples_waveforms=nsw,
filter_order=filter_order,
sample_rate=self.sample_rate,
mask_threshold=mask_threshold,
)
else:
waveforms = None
self.all_waveforms = waveforms
# Read the cluster groups.
logger.debug("Loading the cluster groups.")
self.cluster_groups = {}
if op.exists(filenames['cluster_groups']):
with open(filenames['cluster_groups'], 'r') as f:
reader = csv.reader(f, delimiter='\t')
# Skip the header.
for row in reader:
break
for row in reader:
cluster, group = row
cluster = int(cluster)
self.cluster_groups[cluster] = group
for cluster_id in self.cluster_ids:
if cluster_id not in self.cluster_groups:
self.cluster_groups[cluster_id] = None