def genTrainPairs(self):
tr_pair = []
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3]-index_extent[1])/settings["GRID_LNG"])
ndimy = int((index_extent[2]-index_extent[0])/settings["GRID_LAT"])
user_visit = defaultdict(set)
idx = 0
for entry in csv.reader(open(self.trdata_path)):
sys.stdout.write("\rFINISHED PAIR NUM: %d. " % (idx+1))
sys.stdout.flush()
idx += 1
uid, pid1, pid2 = int(entry[0]), int(entry[1]), int(entry[4])
if pid1 not in user_visit[uid]:
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx, ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
if self.nsample < len(candidate_pois):
for pid in random.sample(set(candidate_pois)-set([pid1]), self.nsample):
tr_pair.append([self.user_ids[uid], self.poi_ids[pid1], self.poi_ids[pid]])
user_visit[uid].add(pid1)
#user_visit[uid].add(pid1)
if pid2 not in user_visit[uid]:
near_grids = getNearGridsForPOI(self.pois_latlng[pid2], ndimx, ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
if self.nsample < len(candidate_pois):
for pid in random.sample(set(candidate_pois)-set([pid2]), self.nsample):
tr_pair.append([self.user_ids[uid], self.poi_ids[pid2], self.poi_ids[pid]])
user_visit[uid].add(pid2)
#user_visit[uid].add(pid2)
print len(tr_pair)
return tr_pair
def recommendationNewPOI(self, submission_path):
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3]-index_extent[1])/settings["GRID_LNG"])
ndimy = int((index_extent[2]-index_extent[0])/settings["GRID_LAT"])
recommendation_result = {}
user_visited = defaultdict(list)
for entry in csv.reader(open(self.trdata_path)):
uid, pid1, pid2 = int(entry[0]), int(entry[1]), int(entry[4])
user_visited[uid].append(pid1)
user_visited[uid].append(pid2)
for i, entry in enumerate(csv.reader(open(self.tedata_path))):
uid, pid1 = int(entry[0]), int(entry[1])
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx, ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
result = []
pois_score = []
for c_pid in set(candidate_pois)-user_visited[uid]:
if self.bias_tag == True:
score = np.dot(self.user_factor[self.user_ids[uid]]+self.query_factor[self.poi_ids[pid1]], self.poi_factor[self.poi_ids[c_pid]])\
+ self.poi_bias[self.poi_ids[c_pid]]
else:
score = np.dot(self.user_factor[self.user_ids[uid]]+self.query_factor[self.poi_ids[pid1]], self.poi_factor[self.poi_ids[c_pid]])
pois_score.append([c_pid, score])
result = sorted(pois_score, key=lambda x:x[1], reverse=True)[:settings["MAX_TOPK"]]
recommendation_result[i] = [pair[0] for pair in result]
sys.stdout.write("\rFINISHED RECOMMENDATION TRIPLE NUM: %d. " % (i+1))
sys.stdout.flush()
write_submission(recommendation_result, submission_path)
def recommendation(self, submission_path):
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3]-index_extent[1])/settings["GRID_LNG"])
ndimy = int((index_extent[2]-index_extent[0])/settings["GRID_LAT"])
recommendation_result = {}
cache_user_poi_score = defaultdict(dict)
for i, entry in enumerate(csv.reader(open(self.tedata_path))):
uid, pid1 = int(entry[0]), int(entry[1])
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx, ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
result = []
pois_score = []
for c_pid in candidate_pois:
if uid in cache_user_poi_score and c_pid in cache_user_poi_score[uid]:
result.append([c_pid, cache_user_poi_score[uid][c_pid]])
else:
if self.bias_tag == True:
score = np.dot(self.user_factor[self.user_ids[uid]],\
self.poi_factor[self.poi_ids[c_pid]])\
+ self.poi_bias[self.poi_ids[c_pid]]
else:
score = np.dot(self.user_factor[self.user_ids[uid]],\
self.poi_factor[self.poi_ids[c_pid]])
pois_score.append([c_pid, score])
cache_user_poi_score[uid][c_pid] = score
result = sorted(pois_score, key=lambda x:x[1], reverse=True)[:settings["MAX_TOPK"]]
recommendation_result[i] = [pair[0] for pair in result]
sys.stdout.write("\rFINISHED PAIR NUM: %d. " % (i+1))
sys.stdout.flush()
write_submission(recommendation_result, submission_path)
def genTrainTriples(self):
tr_triple = []
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3] - index_extent[1]) / settings["GRID_LNG"])
ndimy = int((index_extent[2] - index_extent[0]) / settings["GRID_LAT"])
idx = 0
for entry in csv.reader(open(self.trdata_path)):
sys.stdout.write("\rFINISHED TRIPLE NUM: %d. " % (idx + 1))
sys.stdout.flush()
idx += 1
uid, pid1, pid2 = int(entry[0]), int(entry[1]), int(entry[4])
near_grids = getNearGridsForPOI(self.pois_latlng[pid2], ndimx,
ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
if self.nsample < len(candidate_pois):
for pid in random.sample(
set(candidate_pois) - set([pid2]), self.nsample):
tr_triple.append([
self.user_ids[uid],
self.category_ids[self.poi_category[pid1]],
self.poi_ids[pid1], self.poi_ids[pid2],
self.poi_ids[pid]
])
print len(tr_triple)
return tr_triple
def recommendation(self, submission_path):
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3] - index_extent[1]) / settings["GRID_LNG"])
ndimy = int((index_extent[2] - index_extent[0]) / settings["GRID_LAT"])
recommendation_result = {}
for i, entry in enumerate(csv.reader(open(self.tedata_path))):
uid, pid1 = int(entry[0]), int(entry[1])
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx,
ndimy, True)
pois_score = []
for grididx in near_grids:
for candidate_poi in self.grids_pois[grididx[0]][grididx[1]]:
if candidate_poi in self.per_pois_pop[uid]:
pois_score.append([
candidate_poi,
self.per_pois_pop[uid][candidate_poi]
])
else:
pois_score.append([candidate_poi, 0])
result = sorted(pois_score, key=lambda x: x[1],
reverse=True)[:settings["MAX_TOPK"]]
recommendation_result[i] = [pair[0] for pair in result]
print i
write_submission(recommendation_result, submission_path)
def recommendation(self, submission_path):
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3] - index_extent[1]) / settings["GRID_LNG"])
ndimy = int((index_extent[2] - index_extent[0]) / settings["GRID_LAT"])
recommendation_result = {}
cache_user_poi_score = defaultdict(dict)
for i, entry in enumerate(csv.reader(open(self.tedata_path))):
uid, pid1 = int(entry[0]), int(entry[1])
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx,
ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
result = []
pois_score = []
for c_pid in candidate_pois:
if uid in cache_user_poi_score and c_pid in cache_user_poi_score[
uid]:
result.append([c_pid, cache_user_poi_score[uid][c_pid]])
else:
if self.bias_tag == True:
score = np.dot(self.user_factor[self.user_ids[uid]],\
self.poi_factor[self.poi_ids[c_pid]])\
+ self.poi_bias[self.poi_ids[c_pid]]
else:
score = np.dot(self.user_factor[self.user_ids[uid]],\
self.poi_factor[self.poi_ids[c_pid]])
pois_score.append([c_pid, score])
cache_user_poi_score[uid][c_pid] = score
result = sorted(pois_score, key=lambda x: x[1],
reverse=True)[:settings["MAX_TOPK"]]
recommendation_result[i] = [pair[0] for pair in result]
sys.stdout.write("\rFINISHED PAIR NUM: %d. " % (i + 1))
sys.stdout.flush()
write_submission(recommendation_result, submission_path)
def genTrainPairs(self):
tr_pair = []
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3] - index_extent[1]) / settings["GRID_LNG"])
ndimy = int((index_extent[2] - index_extent[0]) / settings["GRID_LAT"])
user_visit = defaultdict(set)
idx = 0
for entry in csv.reader(open(self.trdata_path)):
sys.stdout.write("\rFINISHED PAIR NUM: %d. " % (idx + 1))
sys.stdout.flush()
idx += 1
uid, pid1, pid2 = int(entry[0]), int(entry[1]), int(entry[4])
if pid1 not in user_visit[uid]:
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx,
ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
if self.nsample < len(candidate_pois):
for pid in random.sample(
set(candidate_pois) - set([pid1]), self.nsample):
tr_pair.append([
self.user_ids[uid], self.poi_ids[pid1],
self.poi_ids[pid]
])
user_visit[uid].add(pid1)
#user_visit[uid].add(pid1)
if pid2 not in user_visit[uid]:
near_grids = getNearGridsForPOI(self.pois_latlng[pid2], ndimx,
ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
if self.nsample < len(candidate_pois):
for pid in random.sample(
set(candidate_pois) - set([pid2]), self.nsample):
tr_pair.append([
self.user_ids[uid], self.poi_ids[pid2],
self.poi_ids[pid]
])
user_visit[uid].add(pid2)
#user_visit[uid].add(pid2)
print len(tr_pair)
return tr_pair
def recommendation(self, submission_path):
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3]-index_extent[1])/settings["GRID_LNG"])
ndimy = int((index_extent[2]-index_extent[0])/settings["GRID_LAT"])
recommendation_result = {}
for i, entry in enumerate(csv.reader(open(self.tedata_path))):
uid, pid1 = int(entry[0]), int(entry[1])
near_grids = getNearGridsForPOI(self.pois_latlng[pid1], ndimx, ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
pois_score = [[poi, self.pois_popularity[poi]] for poi in candidate_pois]
result = sorted(pois_score, key=lambda x:x[1], reverse=True)[:settings["MAX_TOPK"]]
recommendation_result[i] = [pair[0] for pair in result]
print i
write_submission(recommendation_result, submission_path)
def genTrainTriples(self):
tr_triple = []
index_extent = (-90, -180, 90, 180)
ndimx = int((index_extent[3]-index_extent[1])/settings["GRID_LNG"])
ndimy = int((index_extent[2]-index_extent[0])/settings["GRID_LAT"])
idx = 0
for entry in csv.reader(open(self.trdata_path)):
sys.stdout.write("\rFINISHED TRIPLE NUM: %d. " % (idx+1))
sys.stdout.flush()
idx += 1
uid, pid1, pid2 = int(entry[0]), int(entry[1]), int(entry[4])
near_grids = getNearGridsForPOI(self.pois_latlng[pid2], ndimx, ndimy, True)
candidate_pois = []
for grididx in near_grids:
candidate_pois += self.grids_pois[grididx[0]][grididx[1]]
if self.nsample < len(candidate_pois):
for pid in random.sample(set(candidate_pois)-set([pid2]), self.nsample):
tr_triple.append([self.user_ids[uid],
self.category_ids[self.poi_category[pid1]],
self.poi_ids[pid2],
self.poi_ids[pid]])
print len(tr_triple)
return tr_triple