def recommend_with_scores(self, test_device_events, tips):
last_session_events = _get_last_session(test_device_events)
event_to_score = {}
for (group_id, event_id) in self.event_types:
event_to_score[(group_id, event_id)] = 0
for (group_id2, event_id2) in last_session_events.keys():
if (group_id2, event_id2) in self.cp_matrix[(group_id,
event_id)].keys():
event_to_score[(group_id, event_id)] += self.cp_matrix[(
group_id, event_id)][(group_id2, event_id2)]
if (group_id2, event_id2) in self.dp_matrix.keys():
event_to_score[(group_id, event_id)] += self.dp_matrix[(
group_id2, event_id2)][(group_id, event_id)]
sorted_by_score_events = sorted(event_to_score.items(),
key=operator.itemgetter(1),
reverse=True)
tips_to_recommend = {}
for top_event in sorted_by_score_events:
event = top_event[0]
score = top_event[1]
if (event not in test_device_events.keys()) \
and _is_intersection(tips, event_to_tips(event[0], event[1])) > 0:
for tip in event_to_tips(event[0], event[1]):
if tip in tips:
tips_to_recommend[tip] = score
if self.is_logging:
logging.info("RecommenderCo+Dis:recommend: recommendation made.")
return tips_to_recommend
def _generate_recommendation_list(self, recommendations, test_device_events, tips):
recommendation_map = {}
for event, score in recommendations:
if event_to_tips(self.event_types[event][0], self.event_types[event][1]) and \
_is_intersection(tips, event_to_tips(self.event_types[event][0], self.event_types[event][1])) > 0 \
and self.event_types[event] not in test_device_events:
for tip in event_to_tips(self.event_types[event][0], self.event_types[event][1]):
if tip in tips:
recommendation_map[tip] = max(0, score)
return recommendation_map
def process_action(self, event):
if event.device_id not in self.device_to_done_actions.keys():
self.device_to_done_actions[event.device_id] = {}
if (event.event_id, event.ide
) not in self.device_to_done_actions[event.device_id].keys():
self.device_to_done_actions[event.device_id][(
event.event_id, event.ide)] = (event.timestamp, event.count)
else:
_, cnt = self.device_to_done_actions[event.device_id][(
event.event_id, event.ide)]
self.device_to_done_actions[event.device_id][(event.event_id, event.ide)] = \
(event.timestamp, cnt + event.count)
if event.device_id in self.device_to_tips.keys():
for (tip, tip_timestamp,
tip_id) in self.device_to_tips[event.device_id]:
if tip in event_to_tips(group_id=ACTION_INVOKED_GROUP,
event_id=event.event_id):
if event.timestamp - tip_timestamp < PREDICTED_TIME_MILLIS:
with open(
TRAIN_LABELS_POSITIVE_DIR +
self.generate_file_name(
event.device_id, "csv", tip_id),
'w') as fout:
fout.write(tip + "\n")
self.tips_done[tip_id] = True
def process_tip(self, event):
tip_name = event.event_id.split(";")[0]
algo_name = event.event_id.split(";")[1]
if algo_name not in self.all_tips_cnt.keys():
self.all_tips_cnt[algo_name] = 0
self.good_tips_cnt[algo_name] = 0
self.devices_good[algo_name] = {}
self.devices_all[algo_name] = {}
self.all_tips_cnt[algo_name] += 1
self.devices_all[algo_name][event.device_id] = True
is_tip_done_before = False
if event.device_id in self.device_to_done_actions.keys():
for done_action in self.device_to_done_actions[
event.device_id].keys():
event_id, _ = done_action
action_timestamp, _ = self.device_to_done_actions[
event.device_id][done_action]
if tip_name in event_to_tips(group_id=ACTION_INVOKED_GROUP, event_id=event_id) \
and event.timestamp - action_timestamp < FORGET_TIME_DAYS * 24 * 60 * 60 * 1000:
is_tip_done_before = True
break
if not is_tip_done_before:
if event.device_id not in self.device_to_tips.keys():
self.device_to_tips[event.device_id] = []
self.device_to_tips[event.device_id].append(
(tip_name, event.timestamp, algo_name))
def recommend(self, test_device_events, tips):
if self.is_logging:
logging.info(
"RecommenderTopEventWithProbability:recommend: recommendation started."
)
test_device_events = self._filter_old_test_device_events(
test_device_events)
not_done_event_with_prob, all_not_done_top_sum = \
self._get_not_done_events_with_probability(test_device_events, tips)
if self.is_logging:
logging.info(
"RecommenderTopEventWithProbability:recommend: not_done_events computed, "
+ str(len(not_done_event_with_prob)) + " found.")
if len(not_done_event_with_prob) <= 0:
return event_to_tips(self.top_events[0][0][0],
self.top_events[0][0][1])
probs = []
for event in not_done_event_with_prob:
probs.append(event[1] / all_not_done_top_sum)
if self.is_logging:
logging.info(
"RecommenderTopEventWithProbability:recommend: probability for not_done_events computed"
)
events_ids = np.random.choice(range(len(not_done_event_with_prob)),
len(not_done_event_with_prob),
replace=False,
p=probs)
events = np.array(not_done_event_with_prob)[events_ids]
tips_to_recommend = []
for event in events:
for tip in event_to_tips(event[0][0], event[0][1]):
if tip in tips:
tips_to_recommend.append(tip)
if self.is_logging:
logging.info(
"RecommenderTopEventWithProbability:recommend: tip generated.")
return tips_to_recommend
def _split_events_by_ide(events):
ide_to_events = {}
for event in events:
if ide not in ide_to_events.keys():
ide_to_events[ide] = []
if event.group_id == ACTION_INVOKED_GROUP and len(
event_to_tips(event.group_id, event.event_id)) > 0:
ide_to_events[ide].append(
(event.group_id, event.event_id, event.device_id, event.count))
return ide_to_events
def recommend_with_scores(self, test_device_events, tips):
if self.is_logging:
logging.info(
"RecommenderTopEvent:recommend: recommendation started.")
test_device_events = self._filter_old_test_device_events(
test_device_events)
tips_to_recommend = {}
for top_event in self.top_events:
event = top_event[0]
score = top_event[1]
if (event not in test_device_events.keys())\
and _is_intersection(tips, event_to_tips(event[0], event[1])) > 0:
for tip in event_to_tips(event[0], event[1]):
if tip in tips:
tips_to_recommend[tip] = score
if self.is_logging:
logging.info("RecommenderTopEvent:recommend: recommendation made.")
return tips_to_recommend
def _get_not_done_events_with_probability(self, test_device_events, tips):
not_done_event_with_prob = []
all_not_done_top_sum = 0
for top_event in self.top_events:
event = top_event[0]
probability = top_event[1]
if (event not in test_device_events.keys()) \
and _is_intersection(tips, event_to_tips(event[0], event[1])) > 0:
not_done_event_with_prob.append(top_event)
all_not_done_top_sum += probability
return not_done_event_with_prob, all_not_done_top_sum
def process_tip(self, event):
data = {
"tips": [],
"usageInfo": {},
"bucket": event.bucket,
'ideName': event.ide
}
tip_name = event.event_id.split(";")[0]
for t in self.ide_tips[event.ide].keys():
data["tips"].append(t)
is_tip_done_before = False
for done_action in self.device_to_done_actions[event.device_id].keys():
event_id, ide = done_action
max_timestamp, count = self.device_to_done_actions[
event.device_id][done_action]
if tip_name in event_to_tips(group_id=ACTION_INVOKED_GROUP,
event_id=event_id):
is_tip_done_before = True
break
if ide == event.ide:
data["usageInfo"][event_id] = {}
data["usageInfo"][event_id]["usageCount"] = count
data["usageInfo"][event_id][
"lastUsedTimestamp"] = max_timestamp
if not is_tip_done_before:
with open(
TRAIN_EVENTS_DIR + self.generate_file_name(
event.device_id, "json", self.tip_id_cur),
'w') as fout:
json.dump(data, fout)
if event.device_id not in self.device_to_tips.keys():
self.device_to_tips[event.device_id] = []
self.device_to_tips[event.device_id].append(
(tip_name, event.timestamp, str(self.tip_id_cur)))
self.tip_id_cur += 1
def process_action(self, event):
if event.device_id not in self.device_to_done_actions.keys():
self.device_to_done_actions[event.device_id] = {}
if (event.event_id, event.ide
) not in self.device_to_done_actions[event.device_id].keys():
self.device_to_done_actions[event.device_id][(
event.event_id, event.ide)] = (event.timestamp, event.count)
else:
_, cnt = self.device_to_done_actions[event.device_id][(
event.event_id, event.ide)]
self.device_to_done_actions[event.device_id][(event.event_id, event.ide)] = \
(event.timestamp, cnt + event.count)
if event.device_id in self.device_to_tips.keys():
for (tip, tip_timestamp,
algo) in self.device_to_tips[event.device_id]:
if tip in event_to_tips(group_id=ACTION_INVOKED_GROUP,
event_id=event.event_id):
if event.timestamp - tip_timestamp < PREDICTED_TIME_MILLIS:
self.good_tips_cnt[algo] += 1
self.devices_good[algo][event.device_id] = True
