def _plot_cluster(self, bunch):
wave = bunch.data
if wave is None or not wave.size:
return
channel_ids_loc = bunch.channel_ids
n_channels = len(channel_ids_loc)
masks = bunch.get('masks', np.ones((wave.shape[0], n_channels)))
# By default, this is 0, 1, 2 for the first 3 clusters.
# But it can be customized when displaying several sets
# of waveforms per cluster.
n_spikes_clu, n_samples = wave.shape[:2]
assert wave.shape[2] == n_channels
assert masks.shape == (n_spikes_clu, n_channels)
# Find the x coordinates.
t = get_linear_x(n_spikes_clu * n_channels, n_samples)
t = _overlap_transform(t,
offset=bunch.offset,
n=bunch.n_clu,
overlap=self.overlap)
# HACK: on the GPU, we get the actual masks with fract(masks)
# since we add the relative cluster index. We need to ensure
# that the masks is never 1.0, otherwise it is interpreted as
# 0.
masks *= .99999
# NOTE: we add the cluster index which is used for the
# computation of the depth on the GPU.
masks += bunch.index
# Generate the box index (one number per channel).
box_index = _index_of(channel_ids_loc, self.channel_ids)
box_index = np.repeat(box_index, n_samples)
box_index = np.tile(box_index, n_spikes_clu)
assert box_index.shape == (n_spikes_clu * n_channels * n_samples, )
# Generate the waveform array.
wave = np.transpose(wave, (0, 2, 1))
wave = wave.reshape((n_spikes_clu * n_channels, n_samples))
self.waveform_visual.add_batch_data(x=t,
y=wave,
color=bunch.color,
masks=masks,
box_index=box_index,
data_bounds=self.data_bounds)
def _plot_cluster(self, bunch, color=None):
"""Plot one cluster."""
wave = bunch.template # shape: (n_samples, n_channels)
channel_ids_loc = bunch.channel_ids
n_channels_loc = len(channel_ids_loc)
n_samples, nc = wave.shape
assert nc == n_channels_loc
# Find the x coordinates.
t = get_linear_x(n_channels_loc, n_samples)
color = color or self.cluster_colors[bunch.cluster_rel]
assert len(color) == 4
box_index = self._get_box_index(bunch)
return Bunch(
x=t, y=wave.T, color=color, box_index=box_index, data_bounds=self.data_bounds)
def _plot_cluster(self, bunch):
wave = bunch.data
if wave is None or not wave.size:
return
channel_ids_loc = bunch.channel_ids
n_channels = len(channel_ids_loc)
masks = bunch.get('masks', np.ones((wave.shape[0], n_channels)))
# By default, this is 0, 1, 2 for the first 3 clusters.
# But it can be customized when displaying several sets
# of waveforms per cluster.
n_spikes_clu, n_samples = wave.shape[:2]
assert wave.shape[2] == n_channels
assert masks.shape == (n_spikes_clu, n_channels)
# Find the x coordinates.
t = get_linear_x(n_spikes_clu * n_channels, n_samples)
t = _overlap_transform(t,
offset=bunch.offset,
n=bunch.n_clu,
overlap=self.overlap)
# HACK: on the GPU, we get the actual masks with fract(masks)
# since we add the relative cluster index. We need to ensure
# that the masks is never 1.0, otherwise it is interpreted as
# 0.
eps = .001
masks = eps + (1 - 2 * eps) * masks
# NOTE: we add the cluster index which is used for the
# computation of the depth on the GPU.
masks += bunch.index
# Generate the box index (one number per channel).
box_index = _index_of(channel_ids_loc, self.channel_ids)
box_index = np.tile(box_index, n_spikes_clu)
# Find the correct number of vertices depending on the current waveform visual.
if self._current_visual == self.waveform_visual:
# PlotVisual
box_index = np.repeat(box_index, n_samples)
assert box_index.size == n_spikes_clu * n_channels * n_samples
else:
# PlotAggVisual
box_index = np.repeat(box_index, 2 * (n_samples + 2))
assert box_index.size == n_spikes_clu * n_channels * 2 * (
n_samples + 2)
# Generate the waveform array.
wave = np.transpose(wave, (0, 2, 1))
nw = n_spikes_clu * n_channels
wave = wave.reshape((nw, n_samples))
assert self.data_bounds is not None
self._current_visual.add_batch_data(x=t,
y=wave,
color=bunch.color,
masks=masks,
box_index=box_index,
data_bounds=self.data_bounds)
# Waveform axes.
# --------------
# Horizontal y=0 lines.
ax_db = self.data_bounds
a, b = _overlap_transform(np.array([-1, 1]),
offset=bunch.offset,
n=bunch.n_clu,
overlap=self.overlap)
box_index = _index_of(channel_ids_loc, self.channel_ids)
box_index = np.repeat(box_index, 2)
box_index = np.tile(box_index, n_spikes_clu)
hpos = np.tile([[a, 0, b, 0]], (nw, 1))
assert box_index.size == hpos.shape[0] * 2
self.line_visual.add_batch_data(
pos=hpos,
color=self.ax_color,
data_bounds=ax_db,
box_index=box_index,
)
# Vertical ticks every millisecond.
steps = np.arange(np.round(self.wave_duration * 1000))
# A vline every millisecond.
x = .001 * steps
# Scale to [-1, 1], same coordinates as the waveform points.
x = -1 + 2 * x / self.wave_duration
# Take overlap into account.
x = _overlap_transform(x,
offset=bunch.offset,
n=bunch.n_clu,
overlap=self.overlap)
x = np.tile(x, len(channel_ids_loc))
# Generate the box index.
box_index = _index_of(channel_ids_loc, self.channel_ids)
box_index = np.repeat(box_index, x.size // len(box_index))
assert x.size == box_index.size
self.tick_visual.add_batch_data(
x=x,
y=np.zeros_like(x),
data_bounds=ax_db,
box_index=box_index,
)