def _evaluate(project_id, user_id, start_date, end_date, max_gap, min_nodes):
# Obtain neurons that are fully reviewed at the moment
# and to which the user contributed nodes within the date range.
# 1. Find out skeleton_ids towards which the user contributed
# within the date range
ts = (
Treenode.objects.filter(user_id=user_id, creation_time__range=(start_date, end_date))
.values_list("skeleton")
.annotate(Count("skeleton"))
)
# Pick only skeletons for which the user contributed at least min_nodes
skeleton_ids = set(skid for skid, count in ts if count > min_nodes)
if not skeleton_ids:
return None
# Find the subset of fully reviewed skeletons
ts = (
Treenode.objects.filter(skeleton__in=skeleton_ids)
.values_list("skeleton", "reviewer_id")
.annotate(Count("reviewer_id"))
)
review_status = defaultdict(partial(defaultdict, int))
for skid, reviewer_id, count in ts:
review_status[skid][reviewer_id] = count
not_fully_reviewed = set()
for skid, reviewers in review_status.iteritems():
if -1 in reviewers:
not_fully_reviewed.add(skid)
skeleton_ids = skeleton_ids - not_fully_reviewed
if not skeleton_ids:
return None
relations = dict(
Relation.objects.filter(
project_id=project_id, relation_name__in=["presynaptic_to", "postsynaptic_to"]
).values_list("relation_name", "id")
)
# 2. Load each fully reviewed skeleton one at a time
evaluations = {
skid: _evaluate_arbor(user_id, skid, tree, relations, max_gap)
for skid, tree in lazy_load_trees(
skeleton_ids,
("location", "creation_time", "user_id", "reviewer_id", "review_time", "editor_id", "edition_time"),
)
}
# 3. Extract evaluations for the user_id over time
# Each evaluation contains an instance of EpochOps namedtuple, with members:
# 'review_date_range', 'creation_date_range', 'user_node_counts',
# 'splits', 'merges', 'appended', 'node_count'
# The X axis is the last (user) creation date within the review epoch
# The Y axis is multiple, and includes:
# * skeleton_id
# * reviewer_id
# * time of the last node created by the user_id in skeleton_id
# * nodes contributed by the user that were reviewed within the epoch
# * number of nodes missed by the user (which were added by the reviewer)
# * splits onto the user's nodes
# * merges onto the user's nodes
# * additions by the reviewer onto nodes of this user (another form of merges)
# * total number of presynaptic relations of skeleton_id
# * total number of postsynaptic relations of skeleton_id
# * number of presynaptic_to relations created by the reviewer within the review period onto treenodes created by user_id
# * number of postsynaptic_to relations created by the reviewer within the review period onto treenodes created by user_id
# * newer_synapses: number of synapses created by someone else onto treenodes created by user_id, after the creation of the treenode
d = []
for skid, arbor_epoch_ops in evaluations.iteritems():
for epoch_ops in arbor_epoch_ops:
if 0 == epoch_ops.user_node_counts[user_id]:
# user did not contribute at all to this chunk
continue
appended = epoch_ops.appended[user_id]
print appended
d.append(
{
"skeleton_id": skid,
"reviewer_id": epoch_ops.reviewer_id,
"timepoint": epoch_ops.creation_date_range[user_id]["end"].strftime("%Y-%m-%d"),
"n_created_nodes": epoch_ops.user_node_counts[user_id],
"n_nodes": epoch_ops.node_count,
"n_missed_nodes": sum(appended),
"n_splits": epoch_ops.splits[user_id],
"n_merges": epoch_ops.merges[user_id] + len(appended),
"n_pre": epoch_ops.n_pre,
"n_post": epoch_ops.n_post,
"reviewer_n_pre": epoch_ops.reviewer_n_pre.get(user_id, 0),
"reviewer_n_post": epoch_ops.reviewer_n_post.get(user_id, 0),
"newer_pre": epoch_ops.newer_pre_count.get(user_id, 0),
"newer_post": epoch_ops.newer_post_count.get(user_id, 0),
}
)
return d
def _evaluate(project_id, user_id, start_date, end_date, max_gap, min_nodes):
# Obtain neurons that are fully reviewed at the moment
# and to which the user contributed nodes within the date range.
# 1. Find out skeleton_ids towards which the user contributed
# within the date range
ts = Treenode.objects.filter(
user_id=user_id,
creation_time__range = (start_date, end_date)) \
.values_list('skeleton') \
.annotate(Count('skeleton'))
# Pick only skeletons for which the user contributed at least min_nodes
skeleton_ids = set(skid for skid, count in ts if count > min_nodes)
if not skeleton_ids:
return None
# Find the subset of fully reviewed (union without evaluated user) skeletons
review_status = get_review_status(skeleton_ids)
not_fully_reviewed = set()
for skid, status in review_status.iteritems():
if status != 100:
not_fully_reviewed.add(skid)
skeleton_ids = skeleton_ids - not_fully_reviewed
if not skeleton_ids:
return None
# Get review information and organize it by skeleton ID and treenode ID
reviews = defaultdict(lambda: defaultdict(list))
for r in Review.objects.filter(skeleton_id__in=skeleton_ids):
reviews[r.skeleton_id][r.treenode_id].append(r)
# Sort all reviews of all treenodes by review time, most recent first
for skid, tid_to_rs in reviews.iteritems():
for tid, rs in tid_to_rs.iteritems():
rs.sort(key=lambda r: r.review_time)
rs.reverse()
relations = dict(Relation.objects.filter(project_id=project_id, relation_name__in=['presynaptic_to', 'postsynaptic_to']).values_list('relation_name', 'id'))
# 2. Load each fully reviewed skeleton one at a time
evaluations = {skid: _evaluate_arbor(user_id, skid, tree, reviews[skid], relations, max_gap) \
for skid, tree in lazy_load_trees(skeleton_ids, ('location', 'creation_time', 'user_id', 'editor_id', 'edition_time'))}
# 3. Extract evaluations for the user_id over time
# Each evaluation contains an instance of EpochOps namedtuple, with members:
# 'review_date_range', 'creation_date_range', 'user_node_counts',
# 'splits', 'merges', 'appended', 'node_count'
# The X axis is the last (user) creation date within the review epoch
# The Y axis is multiple, and includes:
# * skeleton_id
# * reviewer_id
# * time of the last node created by the user_id in skeleton_id
# * nodes contributed by the user that were reviewed within the epoch
# * number of nodes missed by the user (which were added by the reviewer)
# * splits onto the user's nodes
# * merges onto the user's nodes
# * additions by the reviewer onto nodes of this user (another form of merges)
# * total number of presynaptic relations of skeleton_id
# * total number of postsynaptic relations of skeleton_id
# * number of presynaptic_to relations created by the reviewer within the review period onto treenodes created by user_id
# * number of postsynaptic_to relations created by the reviewer within the review period onto treenodes created by user_id
# * newer_synapses: number of synapses created by someone else onto treenodes created by user_id, after the creation of the treenode
d = []
for skid, arbor_epoch_ops in evaluations.iteritems():
for epoch_ops in arbor_epoch_ops:
if 0 == epoch_ops.user_node_counts[user_id]:
# user did not contribute at all to this chunk
continue
appended = epoch_ops.appended[user_id]
print appended
d.append({'skeleton_id': skid,
'reviewer_id': epoch_ops.reviewer_id,
'timepoint': epoch_ops.creation_date_range[user_id]['end'].strftime('%Y-%m-%d'),
'n_created_nodes': epoch_ops.user_node_counts[user_id],
'n_nodes': epoch_ops.node_count,
'n_missed_nodes': sum(appended),
'n_splits': epoch_ops.splits[user_id],
'n_merges': epoch_ops.merges[user_id] + len(appended),
'n_pre': epoch_ops.n_pre,
'n_post': epoch_ops.n_post,
'reviewer_n_pre': epoch_ops.reviewer_n_pre.get(user_id, 0),
'reviewer_n_post': epoch_ops.reviewer_n_post.get(user_id, 0),
'newer_pre': epoch_ops.newer_pre_count.get(user_id, 0),
'newer_post': epoch_ops.newer_post_count.get(user_id, 0)})
return d
def _evaluate(project_id, user_id, start_date, end_date, max_gap, min_nodes):
# Obtain neurons that are fully reviewed at the moment
# and to which the user contributed nodes within the date range.
# 1. Find out skeleton_ids towards which the user contributed
# within the date range
ts = Treenode.objects.filter(
user_id=user_id,
creation_time__range = (start_date, end_date)) \
.values_list('skeleton') \
.annotate(Count('skeleton'))
# Pick only skeletons for which the user contributed at least min_nodes
skeleton_ids = set(skid for skid, count in ts if count > min_nodes)
if not skeleton_ids:
return None
# Find the subset of fully reviewed skeletons
ts = Treenode.objects.filter(skeleton__in=skeleton_ids) \
.values_list('skeleton', 'reviewer_id') \
.annotate(Count('reviewer_id'))
review_status = defaultdict(partial(defaultdict, int))
for skid, reviewer_id, count in ts:
review_status[skid][reviewer_id] = count
not_fully_reviewed = set()
for skid, reviewers in review_status.iteritems():
if -1 in reviewers:
not_fully_reviewed.add(skid)
skeleton_ids = skeleton_ids - not_fully_reviewed
if not skeleton_ids:
return None
relations = dict(Relation.objects.filter(project_id=project_id, relation_name__in=['presynaptic_to', 'postsynaptic_to']).values_list('relation_name', 'id'))
# 2. Load each fully reviewed skeleton one at a time
evaluations = {skid: _evaluate_arbor(user_id, skid, tree, relations, max_gap) for skid, tree in lazy_load_trees(skeleton_ids, ('location', 'creation_time', 'user_id', 'reviewer_id', 'review_time', 'editor_id', 'edition_time'))}
# 3. Extract evaluations for the user_id over time
# Each evaluation contains an instance of EpochOps namedtuple, with members:
# 'review_date_range', 'creation_date_range', 'user_node_counts',
# 'splits', 'merges', 'appended', 'node_count'
# The X axis is the last (user) creation date within the review epoch
# The Y axis is multiple, and includes:
# * skeleton_id
# * reviewer_id
# * time of the last node created by the user_id in skeleton_id
# * nodes contributed by the user that were reviewed within the epoch
# * number of nodes missed by the user (which were added by the reviewer)
# * splits onto the user's nodes
# * merges onto the user's nodes
# * additions by the reviewer onto nodes of this user (another form of merges)
# * total number of presynaptic relations of skeleton_id
# * total number of postsynaptic relations of skeleton_id
# * number of presynaptic_to relations created by the reviewer within the review period onto treenodes created by user_id
# * number of postsynaptic_to relations created by the reviewer within the review period onto treenodes created by user_id
# * newer_synapses: number of synapses created by someone else onto treenodes created by user_id, after the creation of the treenode
d = []
for skid, arbor_epoch_ops in evaluations.iteritems():
for epoch_ops in arbor_epoch_ops:
if 0 == epoch_ops.user_node_counts[user_id]:
# user did not contribute at all to this chunk
continue
appended = epoch_ops.appended[user_id]
print appended
d.append({'skeleton_id': skid,
'reviewer_id': epoch_ops.reviewer_id,
'timepoint': epoch_ops.creation_date_range[user_id]['end'].strftime('%Y-%m-%d'),
'n_created_nodes': epoch_ops.user_node_counts[user_id],
'n_nodes': epoch_ops.node_count,
'n_missed_nodes': sum(appended),
'n_splits': epoch_ops.splits[user_id],
'n_merges': epoch_ops.merges[user_id] + len(appended),
'n_pre': epoch_ops.n_pre,
'n_post': epoch_ops.n_post,
'reviewer_n_pre': epoch_ops.reviewer_n_pre.get(user_id, 0),
'reviewer_n_post': epoch_ops.reviewer_n_post.get(user_id, 0),
'newer_pre': epoch_ops.newer_pre_count.get(user_id, 0),
'newer_post': epoch_ops.newer_post_count.get(user_id, 0)})
return d