description="Import sample data for recommendation engine")
parser.add_argument('--engine_id', default='test_resource')
parser.add_argument('--url', default="http://localhost:9090")
parser.add_argument('--events_file', default="./cb_events.csv")
parser.add_argument('--queries_file', default="./cb_queries.csv")
parser.add_argument('--user_id', default=None)
parser.add_argument('--user_id_2', default=None)
parser.add_argument('--secret', default=None)
parser.add_argument('--secret_2', default=None)
args = parser.parse_args()
print(args)
events_client = harness.EventsClient(engine_id=args.engine_id,
url=args.url,
threads=5,
qsize=500,
user_id=args.user_id,
user_secret=args.secret)
print(events_client.host)
import_events(events_client, args.events_file)
query_client = harness.QueriesClient(engine_id=args.engine_id,
url=args.url,
threads=5,
qsize=500,
user_id=args.user_id,
user_secret=args.secret)
execute_queries(query_client, args.queries_file)
def run_map_test(data, eventNames, users = None,
primaryEvent = cfg.testing.primary_event,
consider_non_zero_scores = cfg.testing.consider_non_zero_scores_only,
num = 200, K = cfg.testing.map_k,
test = False, harness_url = "http://localhost:9090"):
N_TEST = 2000
user_information = {}
res_data = {}
# Create harness engine for events...
engine_client = harness.EventsClient(
engine_id = cfg.engine_id,
url = harness_url,
threads = 5,
qsize = 500)
import_events(engine_client, data)
logging.info(engine_client.host)
engine_client.close()
# Create query client in harness...
logging.info("Queries for " + cfg.engine_id)
query_client = harness.QueriesClient(
engine_id = cfg.engine_id,
url = harness_url,
threads=5,
qsize=500)
for rec in data:
if rec.event == primaryEvent:
user = rec.entityId
item = rec.targetEntityId
if not users or user in users:
user_information.setdefault(user, []).append(item)
if test:
holdoutUsers = [*user_information.keys()][1:N_TEST]
else:
holdoutUsers = [*user_information.keys()]
prediction = []
ground_truth = []
user_items_cnt = 0.0
users_cnt = 0
for user in tqdm(holdoutUsers):
query = {
"user": user,
"eventNames": eventNames,
"num": num,
}
try:
res = query_client.send_query(query)
# Sort by score then by item name
tuples = sorted([(r["score"], r["item"]) for r in res.json_body['result']], reverse=True)
scores = [score for score, item in tuples]
items = [item for score, item in tuples]
res_data[user] = {
"items": items,
"scores": scores,
}
# Consider only non-zero scores
if consider_non_zero_scores:
if len(scores) > 0 and scores[0] != 0.0:
prediction.append(items)
ground_truth.append(user_information.get(user, []))
user_items_cnt += len(user_information.get(user, []))
users_cnt += 1
else:
prediction.append(items)
ground_truth.append(user_information.get(user, []))
user_items_cnt += len(user_information.get(user, []))
users_cnt += 1
except harness.NotFoundError:
print("Error with user: %s" % user)
return ([metrics.mapk(ground_truth, prediction, k) for k in range(1, K + 1)],
res_data, user_items_cnt / (users_cnt + 0.00001))