import os
import pandas as pd
from datetime import datetime
from geopy import distance
from libcity.data.dataset.trajectory_encoder.abstract_trajectory_encoder import AbstractTrajectoryEncoder
from libcity.utils import parse_time, parse_coordinate
parameter_list = ['dataset', 'min_session_len', 'min_sessions', 'traj_encoder', 'cut_method',
'window_size', 'min_checkins', 'neg_samples']
[docs]class AtstlstmEncoder(AbstractTrajectoryEncoder):
# 这里有问题，需要重新修改
def __init__(self, config):
super().__init__(config)
self.uid = 0
self.location2id = {} # 因为原始数据集中的部分 loc id 不会被使用到因此这里需要重新编码一下
self.loc_id = 0
self.tim_max = 0 # 记录最大的时间编码
if self.config['cut_method'] == 'time_interval':
# 对于以时间窗口切割的轨迹，最大时间编码是已知的
self.tim_max = self.config['window_size'] - 1
self.feature_dict = {'current_loc': 'int', 'loc_neg': 'int',
'current_dis': 'float', 'dis_neg': 'float',
'current_tim': 'float', 'tim_neg': 'float', 'uid': 'int',
'target_loc': 'int', 'target_dis': 'float', 'target_tim': 'float'
}
parameters_str = ''
for key in parameter_list:
if key in self.config:
parameters_str += '_' + str(self.config[key])
self.cache_file_name = os.path.join(
'./libcity/cache/dataset_cache/', 'trajectory_{}.json'.format(parameters_str))
self.data_path = './raw_data/{}/'.format(self.config['dataset'])
self.geo = pd.read_csv(os.path.join(self.data_path, '{}.geo'.format(self.config['dataset'])))
[docs] def encode(self, uid, trajectories, negative_sample):
"""Encoded Method refered to the open source code
https://github.com/drhuangliwei/An-Attention-based-Spatiotemporal-LSTM-Network-for-Next-POI-Recommendation
row index is:
0 1 2 3 4
dyna_id,type,time,entity_id,location
"""
# 直接对 uid 进行重编码
uid = self.uid
self.uid += 1
encoded_trajectories = []
for i, traj in enumerate(trajectories):
current_loc = [] # the checkin poi list
loc_distance = [] # the distance between two checkin
tim_interval = [] # the time interval between two checkin
pre_time = None
pre_lat = None
pre_lon = None
for index, row in enumerate(traj):
loc = row[4]
now_time = parse_time(row[2])
lon, lat = parse_coordinate(self.geo.loc[self.geo['geo_id'] == loc].iloc[0]['coordinates'])
if index == 0:
# for the first checkin, distance and time_interval set to a fixed value
if loc not in self.location2id:
self.location2id[loc] = self.loc_id
self.loc_id += 1
current_loc.append(self.location2id[loc])
tim_interval.append(100) # choose the same fixed value as the reference code
loc_distance.append(1)
else:
if loc not in self.location2id:
self.location2id[loc] = self.loc_id
self.loc_id += 1
current_loc.append(self.location2id[loc])
# the unit of time is second
tim_interval.append(datetime.timestamp(now_time) - datetime.timestamp(pre_time))
loc_distance.append(distance.distance((pre_lat, pre_lon), (lat, lon)).kilometers)
pre_time = now_time
pre_lat = lat
pre_lon = lon
# generate negative samples' current_loc loc_distance and tim_interval
neg_loc = []
neg_distance = []
neg_time = []
# the final checkin will be target (positive sample), so use the second last to cal neg
row = traj[-2]
loc = row[4]
pre_lon, pre_lat = parse_coordinate(self.geo.loc[self.geo['geo_id'] == loc].iloc[0]['coordinates'])
for neg in negative_sample[i]:
neg_lon, neg_lat = parse_coordinate(self.geo.loc[self.geo['geo_id'] == neg].iloc[0]['coordinates'])
if neg not in self.location2id:
self.location2id[neg] = self.loc_id
self.loc_id += 1
neg_loc.append(self.location2id[neg])
neg_time.append(tim_interval[-1]) # use target's time interval as the neg sample's
neg_distance.append(distance.distance((neg_lat, neg_lon), (pre_lat, pre_lon)).kilometers)
trace = []
target_loc = current_loc[-1]
target_dis = loc_distance[-1]
target_tim = tim_interval[-1]
trace.append(current_loc[:-1])
trace.append(neg_loc)
trace.append(loc_distance[:-1])
trace.append(neg_distance)
trace.append(tim_interval[:-1])
trace.append(neg_time)
trace.append(uid)
trace.append(target_loc)
trace.append(target_dis)
trace.append(target_tim)
encoded_trajectories.append(trace)
return encoded_trajectories
[docs] def gen_data_feature(self):
loc_pad = self.loc_id
self.pad_item = {
'current_loc': loc_pad,
'current_dis': 0.0,
'current_tim': 0.0
}
self.data_feature = {
'loc_size': self.loc_id + 1,
'uid_size': self.uid,
'loc_pad': loc_pad
}