def redo(self):
"""Redo the last cluster assignment operation.
Returns
-------
up : UpdateInfo instance of the changes done by this operation.
"""
# Go forward in the stack, and retrieve the new assignment.
item = self._undo_stack.forward()
if item is None:
# No redo has been performed: abort.
return
# NOTE: the undo_state object is only returned when undoing.
# It represents data associated to the state
# *before* the action. What might be more useful would be the
# undo_state object of the next item in the list (if it exists).
spike_ids, cluster_ids, undo_state = item
assert spike_ids is not None
# We apply the new assignment.
up = self._do_assign(spike_ids, cluster_ids)
up.history = 'redo'
emit('cluster', self, up)
return up
def undo(self):
"""Undo the last cluster assignment operation.
Returns
-------
up : UpdateInfo instance of the changes done by this operation.
"""
_, _, undo_state = self._undo_stack.back()
# Retrieve the initial spike_cluster structure.
spike_clusters_new = self._spike_clusters_base.copy()
# Loop over the history (except the last item because we undo).
for spike_ids, cluster_ids, _ in self._undo_stack:
# We update the spike clusters accordingly.
if spike_ids is not None:
spike_clusters_new[spike_ids] = cluster_ids
# What are the spikes affected by the last changes?
changed = np.nonzero(self._spike_clusters != spike_clusters_new)[0]
clusters_changed = spike_clusters_new[changed]
up = self._do_assign(changed, clusters_changed)
up.history = 'undo'
# Add the undo_state object from the undone object.
up.undo_state = undo_state
emit('cluster', self, up)
return up
def on_mouse_click(self, e):
"""Select a time from the amplitude view to display in the trace view."""
# from pdb import set_trace
# set_trace()
if 'Shift' in e.modifiers:
mouse_pos = self.canvas.panzoom.window_to_ndc(e.pos)
time = Range(NDC, self.data_bounds).apply(mouse_pos)[0][0]
emit('select_time', self, time)
def merge(self, cluster_ids, to=None):
"""Merge several clusters to a new cluster.
Parameters
----------
cluster_ids : array-like
List of clusters to merge.
to : integer
The id of the new cluster. By default, this is `new_cluster_id()`.
Returns
-------
up : UpdateInfo instance
"""
if not _is_array_like(cluster_ids):
raise ValueError(
"The first argument should be a list or an array.")
cluster_ids = sorted(cluster_ids)
if not set(cluster_ids) <= set(self.cluster_ids):
raise ValueError("Some clusters do not exist.")
# Find the new cluster number.
if to is None:
to = self.new_cluster_id()
if to < self.new_cluster_id():
raise ValueError(
"The new cluster numbers should be higher than {0}.".format(
self.new_cluster_id()))
# NOTE: we could have called self.assign() here, but we don't.
# We circumvent self.assign() for performance reasons.
# assign() is a relatively costly operation, whereas merging is a much
# cheaper operation.
# Find all spikes in the specified clusters.
spike_ids = _spikes_in_clusters(self.spike_clusters, cluster_ids)
up = self._do_merge(spike_ids, cluster_ids, to)
undo_state = emit('request_undo_state', self, up)
# Add to stack.
self._undo_stack.add((spike_ids, [to], undo_state))
emit('cluster', self, up)
return up
def assign(self, spike_ids, spike_clusters_rel=0):
"""Make new spike cluster assignments.
Parameters
----------
spike_ids : array-like
List of spike ids.
spike_clusters_rel : array-like
Relative cluster ids of the spikes in `spike_ids`. This
must have the same size as `spike_ids`.
Returns
-------
up : UpdateInfo instance
Note
----
`spike_clusters_rel` contain *relative* cluster indices. Their values
don't matter: what matters is whether two give spikes
should end up in the same cluster or not. Adding a constant number
to all elements in `spike_clusters_rel` results in exactly the same
operation.
The final cluster ids are automatically generated by the `Clustering`
class. This is because we must ensure that all modified clusters
get brand new ids. The whole library is based on the assumption that
cluster ids are unique and "disposable". Changing a cluster always
results in a new cluster id being assigned.
If a spike is assigned to a new cluster, then all other spikes
belonging to the same cluster are assigned to a brand new cluster,
even if they were not changed explicitely by the `assign()` method.
In other words, the list of spikes affected by an `assign()` is almost
always a strict superset of the `spike_ids` parameter. The only case
where this is not true is when whole clusters change: this is called
a merge. It is implemented in a separate `merge()` method because it
is logically much simpler, and faster to execute.
"""
assert not isinstance(spike_ids, slice)
# Ensure `spike_clusters_rel` is an array-like.
if not hasattr(spike_clusters_rel, '__len__'):
spike_clusters_rel = spike_clusters_rel * np.ones(len(spike_ids),
dtype=np.int64)
spike_ids = _as_array(spike_ids)
if len(spike_ids) == 0:
return UpdateInfo()
assert len(spike_ids) == len(spike_clusters_rel)
assert spike_ids.min() >= 0
assert spike_ids.max() < self._n_spikes, "Some spikes don't exist."
# Normalize the spike-cluster assignment such that
# there are only new or dead clusters, not modified clusters.
# This implies that spikes not explicitly selected, but that
# belong to clusters affected by the operation, will be assigned
# to brand new clusters.
spike_ids, cluster_ids = _extend_assignment(spike_ids,
self._spike_clusters,
spike_clusters_rel,
self.new_cluster_id())
up = self._do_assign(spike_ids, cluster_ids)
undo_state = emit('request_undo_state', self, up)
# Add the assignment to the undo stack.
self._undo_stack.add((spike_ids, cluster_ids, undo_state))
emit('cluster', self, up)
return up