def delete_testset(testset_id):
testset = db.query(TestSet).get(testset_id)
if not testset:
abort(404)
testset.clean_up()
db.delete(testset)
db.commit()
return '', 204
def delete_embedding(embedding_id):
embedding = db.query(Embedding).get(embedding_id)
if not embedding:
abort(404)
embedding.clean_up()
db.delete(embedding)
db.commit()
return '', 204
def delete_result(embedding_id, testset_id):
result = db.query(Result).get((embedding_id, testset_id))
if not result:
abort(404)
# Delete its testing_job first.
db.delete(result.testing_job)
db.delete(result)
db.commit()
return '', 204
def delete_testing_job(testing_job_id):
testing_job = db.query(TestingJob).get(testing_job_id)
if not testing_job:
abort(404)
# If it has any result associated, delete it.
result = db.query(Result).get((
testing_job.embedding_id,
testing_job.testset_id
))
if result:
db.delete(result)
if testing_job.task_id:
celery_app.control.revoke(testing_job.task_id, terminate=True)
db.delete(testing_job)
db.commit()
return '', 204
def create_testing_job():
data = request.get_json(force=True)
if 'embedding_id' not in data or 'testset_id' not in data:
abort(400)
embedding_id = data['embedding_id']
testset_id = data['testset_id']
if not (isinstance(embedding_id, int) or isinstance(testset_id, int)):
return jsonify({
'message': "At least one ID must be specified",
'error': 'Bad Request'
}), 400
# Build a list of embeddings and testsets to test.
embeddings = []
testsets = []
if isinstance(embedding_id, int):
embedding = db.query(Embedding).get(embedding_id)
if not embedding:
abort(404)
embeddings.append(embedding)
if isinstance(testset_id, int):
testset = db.query(TestSet).get(testset_id)
testsets.append(testset)
elif testset_id == 'full':
testsets.extend(db.query(TestSet).all())
elif testset_id == 'missing':
existing = db.query(TestSet.id).join(Result).join(Embedding)\
.filter(Embedding.id == embedding_id)
query = db.query(TestSet).filter(~TestSet.id.in_(existing))
testsets.extend(query.all())
elif isinstance(testset_id, int):
testset = db.query(TestSet).get(testset_id)
if not testset:
abort(404)
testsets.append(testset)
if isinstance(embedding_id, int):
embedding = db.query(Embedding).get(embedding_id)
embeddings.append(embedding)
elif embedding_id == 'full':
embeddings.extend(db.query(Embedding).all())
elif embedding_id == 'missing':
existing = db.query(Embedding.id).join(Result).join(TestSet)\
.filter(TestSet.id == testset_id)
query = db.query(Embedding).filter(~Embedding.id.in_(existing))
embeddings.extend(query.all())
# Make sure there are no Nones (i.e. all the models exist).
if any([emb is None for emb in embeddings]):
abort(404)
if any([ts is None for ts in testsets]):
abort(404)
# Make sure the embeddings are trained already.
embeddings = filter(lambda e: e.status == 'TRAINED', embeddings)
# For each pair <embedding, testset>, create the necessary TestingJob,
# deleting it first if it already exists. Also delete associated results.
jobs = []
for embedding in embeddings:
for testset in testsets:
job = db.query(TestingJob)\
.filter_by(embedding=embedding, testset=testset)\
.first()
if job and job.status in ['PENDING', 'PROGRESS']:
# Only overwrite TestingJobs that have already run. If it's
# still pending or running right now, we want to keep it.
continue
elif job:
for result in job.results.all():
db.delete(result)
db.delete(job)
job = TestingJob(testset=testset, embedding=embedding)
jobs.append(job)
db.add(job)
db.commit()
for job in jobs:
test.delay(job.id)
return jsonify(data={'testing_job_id': [job.id for job in jobs]})