def insert_citation(review_id, ref_id, citation_d):
citation = model.Citation()
citation.project_id = review_id
# Ensure that ref_id is an integer, drop it otherwise.
# try:
# ref_id = int(ref_id)
# except:
# ref_id = None
citation.refman = ref_id
pmid = citation_d['pmid']
citation.pmid = pmid if (pmid is not None and pmid != '') else 0
# we truncate the citation if it's too long!
citation.title = citation_d['title'][:MAX_TITLE_LENGTH] if \
citation_d['title'] is not None else "(no title found)"
citation.abstract = citation_d['abstract'][:9980] if \
citation_d['abstract'] is not None else None
citation.authors = " and ".join(citation_d['authors'])
citation.keywords = ','.join(citation_d['keywords'])
citation.journal = citation_d['journal']
model.Session.add(citation)
model.Session.commit()
return citation.id
def test_delete_citation(self):
""" Deleting Citation entry cascades
Verify that entries in CitationTask table are destroyed when
corresponding citation and/or task is deleted
"""
# Create citation and task objects
c1 = model.Citation()
c2 = model.Citation()
t1 = model.Task()
t2 = model.Task()
# Append tasks to citation, incidentally this verifies that
# the relationship are properly set.
c1.tasks.append(t1)
c1.tasks.append(t2)
c2.tasks.append(t2)
# Persist the changes.
Session.add(c1)
Session.add(c2)
Session.commit()
# Verify first that the entries in table CitationTask actually exist
assert len(Session.query(model.citations_tasks_table).all()) == 3
# Finally remove one of the citations and check cascade
Session.delete(c1)
Session.commit()
assert len(Session.query(model.citations_tasks_table).all()) == 1
# Do the same for when the task is removed
Session.delete(t2)
assert len(Session.query(model.citations_tasks_table).all()) == 0
def _create_reviews(p_id, iter_size, which_iter):
lock_file_path = join(dirname(abspath(__file__)), '_delete_lock.lck')
if not isfile(lock_file_path):
Session.query(
model.Citation).filter(model.Citation.project_id != p_id).delete()
Session.query(
model.Label).filter(model.Label.project_id != p_id).delete()
Session.commit()
open(lock_file_path, 'w+').close()
u_id = 2629
k_init = 400
c_count = len(
Session.query(model.Citation).filter_by(project_id=p_id).all())
k_inc = 100
for itercount in range(iter_size * which_iter,
iter_size * which_iter + iter_size):
### THIS is the code for one run of the experiment
## labeled citation counter
labeled_citation_counter = 0
labels = Session.query(model.Label).filter_by(project_id=p_id).all()
user = Session.query(model.User).filter_by(id=u_id).first()
citations = Session.query(
model.Citation).filter_by(project_id=p_id).all()
print len(citations)
c_count = len(citations)
r_sample = defaultdict(list)
sample_indexes = sample(range(c_count), k_init)
C_r = []
for ii in sample_indexes:
C_r.append(citations[ii])
for cc in C_r:
for ll in Session.query(model.Label).filter_by(
project_id=p_id).filter_by(study_id=cc.id).all():
r_sample[ll.study_id].append(ll)
new_review = model.Project()
new_review.leaders.append(user)
new_review.initial_round_size = 0
new_review.tag_privacy = True
Session.add(new_review)
Session.flush()
state_dict = defaultdict(int)
citation_dict = {}
for c in citations:
citation = model.Citation()
citation.project_id = new_review.id
citation.title = c.title
citation.abstract = c.abstract
citation.keywords = c.keywords
citation.refman = c.refman
model.Session.add(citation)
Session.flush()
citation_dict[citation.id] = c.id
if c.id in r_sample:
labeled_citation_counter += 1
state_dict[citation.id] = 1
for t in r_sample[c.id]:
label = model.Label()
label.project_id = new_review.id
label.study_id = citation.id
label.label = t.label
model.Session.add(label)
print new_review.id
Session.commit()
## i is a counter for the current increment
i = 0
while True:
## we want to change the increment size if there are a certain number of citations is labeled
#if labeled_citation_counter > 15000:
# k_inc = 2000
#elif labeled_citation_counter > 5000:
# k_inc = 1000
#else:
# k_inc = 500
r_sample = defaultdict(list)
print "EXPERIMENT NO: " + str(itercount)
make_predictions(new_review.id)
######################## here's where I record the results
preds_for_review = Session.query(model.Prediction).filter(
model.Prediction.project_id == new_review.id).all()
path_to_preds_out = os.path.join(
"_exports",
"predictions_%d_%d_of_%d.csv" % (p_id, i, itercount))
with open(path_to_preds_out, 'w+') as fout:
csv_out = csv.writer(fout)
preds_file_headers = [
"citation_id", "refman", "title",
"predicted p of being relevant",
"'hard' screening prediction*", "state"
]
csv_out.writerow(preds_file_headers)
sorted_preds = sorted(preds_for_review,
key=lambda x: x.predicted_probability,
reverse=True)
for pred in sorted_preds:
citation = Session.query(model.Citation).filter(
model.Citation.id == pred.study_id).first()
#citation = self._get_citation_from_id(pred.study_id)
citation_title = citation.title.encode('ascii', 'ignore')
row_str = [
citation.id, citation.refman, citation_title,
pred.predicted_probability, pred.prediction,
state_dict[citation.id]
]
csv_out.writerow(row_str)
######################### ---------------------------
i += 1
if labeled_citation_counter >= c_count:
break
P_a = []
for pa in Session.query(model.Prediction).filter_by(
project_id=new_review.id).order_by(
model.Prediction.predicted_probability.desc()).all():
if state_dict[pa.study_id] == 0:
P_a.append(pa)
if len(P_a) == k_inc:
break
if len(P_a) == 0:
print "~~~NO PREDS!!!"
ccc = [
label
for label in Session.query(model.Citation.id).filter_by(
project_id=new_review.id).filter(
~model.Citation.labels.any()).limit(k_inc)
]
print len(ccc)
for cc in ccc:
labeled_citation_counter += 1
state_dict[cc.id] = 1
for ll in Session.query(model.Label).filter_by(
study_id=citation_dict[cc.id]).all():
label = model.Label()
label.project_id = new_review.id
label.study_id = cc.id
label.label = ll.label
model.Session.add(label)
else:
for pp in P_a:
labeled_citation_counter += 1
state_dict[pp.study_id] = 2
for ll in Session.query(
model.Label).filter_by(project_id=p_id).filter_by(
study_id=citation_dict[pp.study_id]).all():
label = model.Label()
label.project_id = new_review.id
label.study_id = pp.study_id
label.label = ll.label
model.Session.add(label)
Session.commit()
print len(
Session.query(
model.Label).filter_by(project_id=new_review.id).all())
return