def train_model(user_name, table_name, start, end, pos, neg, step): # Train the model for a user using the data in %table_name% between %start% and %end% # Then store the model and return the model # Format of start and end like '20180309' clf_name = get_clf_name(user_name, pos, neg, start) cnxs = localization.read_local_data(start, end, table_name) intervals, ap_list, cnx_density, ap_freq = localization.create_intervals(cnxs) data = localization.create_training_data(intervals, ap_list, pos, neg, cnx_density) clf = localization.semi_supervised_learning(data, step) model = Model(clf, ap_list, cnx_density, ap_freq) pickle.dump(model, open(clf_name, 'wb')) return model
def coarse_baseline(query_time, user, table):
# Return 0 means outside, 1 means inside
# The second string means the access point if 1 is returned.
# Assume the location data for a user exists and load all events in the queried day
date_str = str(query_time.date()).replace('-', '')
all_events = localization.read_local_data(date_str, date_str, table, True,
user)
if len(all_events) == 0 or len(all_events) == 1:
print(
'Query for %s at %s: Not enough connection activities on that day.'
% (user, query_time))
return 0, ''
n = len(all_events)
if query_time < all_events[0][0] or query_time >= all_events[n - 1][0]:
print(
f'B: Query time {query_time} for {user} is outside the first event {all_events[0]} and the last event {all_events[-1]}'
)
return 0, ''
# compute p as the index of predecessor event to query_time
left = 0
right = n - 1
while left < right:
mid = (left + right + 1) // 2
if query_time < all_events[mid][0]:
right = mid - 1
else:
left = mid
p = left
# online or not decision based on the duration between two events
if p == n - 1:
p = n - 2
duration = (all_events[p + 1][0] - all_events[p][0]).total_seconds()
if duration < 3600:
return 1, all_events[p][1]
else:
return 0, ''
def train_bl_model(user_name,
table_name,
start,
end,
pos,
neg,
step,
validity,
simu=False):
# Train the model for building level localization
clf_name = get_bl_clf_name(user_name, pos, neg, start, end, validity)
all_events = localization.read_local_data(start, end, table_name, simu,
user_name)
gaps, cnx_density, ap_cnx_count = localization.create_gaps(
all_events, validity)
data = localization.create_training_data(gaps, list(ap_cnx_count.keys()),
pos, neg, cnx_density)
clf = localization.semi_supervised_learning(data, step)
model = Model(clf, list(ap_cnx_count.keys()), cnx_density, ap_cnx_count)
pickle.dump(model, open(clf_name, 'wb'))
return model
def answer_query(query_time, user_name, table_name, start=start_date, end=end_date,
pos=pos_threshold, neg=neg_threshold, step=step_size):
# Check whether the model exists for a user, if not generate one, if does, load one
# query_time: datetime type python
date_str = str(query_time.date()).replace('-', '')
all_entries = localization.read_local_data(date_str, date_str, table_name)
if len(all_entries) == 0 or len(all_entries) == 1:
print('Query %s for %s: No connection activities on that day.' % (query_time, user_name))
return 0, ''
intervals, _, _, _ = localization.create_intervals(all_entries)
interval = localization.find_interval(query_time, intervals)
if interval is None:
print('Query %s for %s: Query time before the first connection or after the last on that day.' % (
query_time, user_name))
return 0, ''
try:
f = open(get_clf_name(user_name, pos, neg, start), 'rb')
model = pickle.load(f)
except FileNotFoundError:
model = train_model(user_name, table_name, start, end, pos, neg, step)
state, ap = localization.predict_an_interval(interval, model.ap_list, model.clf, model.density, model.ap_freq)
return state, ap
def train_rl_model(user_name,
table_name,
start,
end,
pos,
neg,
step,
validity,
freq,
simu=False):
# Train the model for region level localization
clf_name = get_rl_clf_name(user_name, pos, neg, start, end, validity, freq)
all_events = localization.read_local_data(start, end, table_name, simu,
user_name)
gaps, cnx_density, ap_cnx_count = localization.create_gaps(
all_events, validity)
data = localization.create_region_train_data(gaps,
list(ap_cnx_count.keys()),
pos, neg, cnx_density,
freq_th, ap_cnx_count)
clf = localization.semi_supervised_learning_new(data, step)
pickle.dump(clf, open(clf_name, 'wb'))
return clf
def answer_query_new(query_time,
user_name,
table_name,
start=start_date,
end=end_date,
pos=pos_threshold,
neg=neg_threshold,
validity=validity_second,
pos_r=pos_r_th,
neg_r=neg_r_th,
freq=freq_th,
fast=False,
step=step_size,
simu=False):
# Assume the location data for a user exists and load all events in the queried day
date_str = str(query_time.date()).replace('-', '')
read_start = process_time()
all_entries = localization.read_local_data(date_str, date_str, table_name,
simu, user_name)
read_end = process_time()
print('data read time: %.2f ms' % (1000 * (read_end - read_start), ))
run_start = process_time()
if len(all_entries) == 0 or len(all_entries) == 1:
if len(all_entries) == 1 and all_entries[0][0] == query_time:
return 1, all_entries[0][1]
print(
'Query for %s at %s: Not enough connection activities on that day.'
% (user_name, query_time))
return 0, ''
# First check whether online or not, if online, work is done
state, ap = localization.online_or_not(query_time, all_entries, validity,
user_name)
if state == 0 or state == 1:
print('Execution time: %.2f ms' %
((process_time() - run_start) * 1000, ))
return state, ap
# Further analysis is needed. The semi-supervised learning method comes. First building level model.
try:
f = open(get_bl_clf_name(user_name, pos, neg, start, end, validity),
'rb')
bl_model = pickle.load(f)
except FileNotFoundError:
bl_model = train_bl_model(user_name, table_name, start, end, pos, neg,
step, validity)
# Then region level model
if fast:
rl_clf = 0
else:
try:
f = open(
get_rl_clf_name(user_name, pos_r, neg_r, start, end, validity,
freq), 'rb')
rl_clf = pickle.load(f)
except FileNotFoundError:
rl_clf = train_rl_model(user_name, table_name, start, end, pos_r,
neg_r, step, validity, freq)
state, ap = localization.predict_a_gap(state, bl_model, rl_clf, fast)
print('Execution time: %.2f ms' % ((process_time() - run_start) * 1000, ))
return state, ap