DiffMM/dataset.py at main · decisionintelligence/DiffMM · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
import os.path
import pickle

import torch
import numpy as np
import tqdm
import random
from torch.utils.data import Dataset
from torch.nn.utils.rnn import pad_sequence

from utils.parse_traj import ParseMMTraj
from utils.model_utils import get_normalized_t, gps2grid


class MMDataset(Dataset):
    def __init__(self, rn, traj_dir, mbr, args, mode, device):
        self.rn = rn
        self.mbr = mbr
        self.args = args
        self.seg_size = rn.valid_edge_cnt
        self.device = device

        self.grid_size = args.grid_size
        self.time_span = args.time_span
        self.keep_ratio = args.keep_ratio

        # self.src_trajs = None
        self.src_norm_gps_seq, self.src_segs_id, self.src_segs_feat = [], [], []
        self.trg_rid = []

        self.mode = mode

        self.get_data(traj_dir)

    def __len__(self):
        # return len(self.src_trajs)
        return len(self.src_norm_gps_seq)

    def __getitem__(self, idx):
        # src_segs_id begins from 1, trg_rid begins from 0
        trg_rid = self.trg_rid[idx].clone().detach().to(self.device)
        trg_onehot = torch.zeros((trg_rid.size(0), self.seg_size), device=self.device)
        for i, rid in enumerate(trg_rid):
            trg_onehot[i, rid] = 1

        return (self.src_norm_gps_seq[idx].clone().detach().to(self.device), self.trg_rid[idx].clone().detach().to(self.device), trg_onehot.clone().detach().to(self.device),
                self.src_segs_id[idx].copy(), self.src_segs_feat[idx].copy())
        # src_traj = self.src_trajs[idx]
        #
        # length = len(src_traj.pt_list)
        #
        # keep_index = [0] + sorted(random.sample(range(1, length - 1), int((length - 2) * self.keep_ratio))) + [
        #     length - 1]
        #
        # src_list = np.array(src_traj.pt_list, dtype=object)
        # src_list = src_list[keep_index].tolist()
        #
        # src_gps_seq, src_norm_gps_seq, trg_rid = self.get_src_seq(src_list)
        #
        # src_segs = self.rn.get_src_segs(src_gps_seq, self.args.search_dist, self.args.beta)
        # segs_id, segs_feat = self.get_segs_feats(src_segs)
        # src_norm_gps_seq = torch.tensor(src_norm_gps_seq)
        #
        # trg_onehot = torch.zeros((length, self.seg_size))
        # for i, rid in enumerate(trg_rid):
        #     trg_onehot[i, rid] = 1
        # trg_rid = torch.tensor(trg_rid)
        # return src_norm_gps_seq, trg_rid, trg_onehot, segs_id, segs_feat

    def get_data(self, traj_dir):
        parser = ParseMMTraj(self.rn)

        if self.mode == 'train':
            src_file = os.path.join(traj_dir, 'traj_train.txt')
        elif self.mode == 'valid':
            src_file = os.path.join(traj_dir, 'traj_valid.txt')
        elif self.mode == 'test':
            src_file = os.path.join(traj_dir, 'traj_test.txt')
        else:
            raise NotImplementedError

        pkl_path = os.path.join(traj_dir, self.mode + '_data_' + str(self.args.search_dist) + '_' + str(self.keep_ratio) + '.pkl')
        if os.path.exists(pkl_path):
            with open(pkl_path, 'rb') as fp:
                self.src_norm_gps_seq, self.trg_rid, self.src_segs_id, self.src_segs_feat = pickle.load(fp)

            # if self.mode == 'train':
            #     self.src_norm_gps_seq, self.trg_rid, self.src_segs_id, self.src_segs_feat = self.src_norm_gps_seq[:128000], self.trg_rid[:128000], self.src_segs_id[:128000], self.src_segs_feat[:128000]

            if self.mode == 'valid':
                self.src_norm_gps_seq, self.trg_rid, self.src_segs_id, self.src_segs_feat = self.src_norm_gps_seq[:10000], self.trg_rid[:10000], self.src_segs_id[:10000], self.src_segs_feat[:10000]

            # if self.mode == 'test':
            #     self.src_norm_gps_seq, self.trg_rid, self.src_segs_id, self.src_segs_feat = self.src_norm_gps_seq[:1], self.trg_rid[:1], self.src_segs_id[:1], self.src_segs_feat[:1]
            #     pickle.dump(self.trg_rid[:1], open('rid.pkl', 'wb'))
            #     pickle.dump(self.src_norm_gps_seq[:1], open('gps.pkl', 'wb'))
        else:
            src_trajs = parser.parse(src_file, is_target=True)

            for src_traj in tqdm.tqdm(src_trajs, desc='traj num'):
                length = len(src_traj.pt_list)
                keep_index = [0] + sorted(random.sample(range(1, length - 1), int((length - 2) * self.keep_ratio))) + [
                    length - 1]

                src_list = np.array(src_traj.pt_list, dtype=object)
                src_list = src_list[keep_index].tolist()

                src_gps_seq, src_norm_gps_seq, trg_rid = self.get_src_seq(src_list)

                src_segs = self.rn.get_src_segs(src_gps_seq, self.args.search_dist, self.args.beta)
                segs_id, segs_feat = self.get_segs_feats(src_segs)
                src_norm_gps_seq = torch.tensor(src_norm_gps_seq)

                # trg_onehot = torch.zeros((length, self.seg_size))
                # for i, rid in enumerate(trg_rid):
                #     trg_onehot[i, rid] = 1
                trg_rid = torch.tensor(trg_rid)

                self.src_norm_gps_seq.append(src_norm_gps_seq)
                self.trg_rid.append(trg_rid)
                # self.trg_onehot.append(trg_onehot)
                self.src_segs_id.append(segs_id)
                self.src_segs_feat.append(segs_feat)

            with open(pkl_path, 'wb') as fp:
                pickle.dump((self.src_norm_gps_seq, self.trg_rid, self.src_segs_id, self.src_segs_feat), fp)

            if self.mode == 'valid':
                self.src_norm_gps_seq, self.trg_rid, self.src_segs_id, self.src_segs_feat = self.src_norm_gps_seq[:10000], self.trg_rid[:10000], self.src_segs_id[:10000], self.src_segs_feat[:10000]

    def get_src_seq(self, ds_pt_list):
        ls_gps_seq = []
        ls_norm_gps_seq = []
        mm_eids = []
        mm_onehot_eids = []
        first_pt = ds_pt_list[0]
        last_pt = ds_pt_list[-1]
        time_interval = self.time_span

        for ds_pt in ds_pt_list:
            ls_gps_seq.append([ds_pt.lat, ds_pt.lng])

            normed_lat = (ds_pt.lat - self.rn.minLat) / (self.rn.maxLat - self.rn.minLat)
            normed_lng = (ds_pt.lng - self.rn.minLon) / (self.rn.maxLon - self.rn.minLon)
            t = get_normalized_t(first_pt, ds_pt, time_interval)
            ls_norm_gps_seq.append([normed_lat, normed_lng, t])

            mm_eids.append(ds_pt.data['candi_pt'].eid)

        return ls_gps_seq, ls_norm_gps_seq, mm_eids

    def get_segs_feats(self, ls_seg):
        seg_id = []
        seg_feat = []

        for segs in ls_seg:
            tmp_id = []
            tmp_feat = []
            for seg in segs:
                tmp_id.append(seg.eid + 1)
                tmp_feat.append([seg.err_weight, seg.cosv, seg.cosv_pre, seg.cosf, seg.cosl, seg.cos1, seg.cos2, seg.cos3, seg.cosp])

            seg_id.append(tmp_id)
            seg_feat.append(tmp_feat)

        return seg_id, seg_feat


def collate_fn(data, device):
    norm_gps_seq, trg_rid, trg_onehot, segs_id, segs_feat = zip(*data)
    lengths = [len(seq) for seq in norm_gps_seq]
    segs_len = [len(seg) for seq in segs_id for seg in seq]
    max_len = max(segs_len)
    segs_id = list(segs_id)
    segs_feat = list(segs_feat)
    segs_mask = []
    for i, segs_seq in enumerate(segs_id):
        tmp_mask = []
        for j, segs in enumerate(segs_id[i]):
            tmp_mask.append([1] * len(segs) + [0] * (max_len - len(segs)))
            segs_id[i][j] = torch.cat((torch.tensor(segs_id[i][j]), torch.zeros(max_len - len(segs))), dim=-1).tolist()
        segs_id[i] = torch.tensor(segs_id[i], device=device)
        segs_mask.append(torch.tensor(tmp_mask, device=device))

    feat_dim = len(segs_feat[0][0][0])
    for i, feats_seq in enumerate(segs_feat):
        for j, feats in enumerate(feats_seq):
            segs_feat[i][j] = torch.cat((torch.tensor(segs_feat[i][j]), torch.zeros((max_len-len(feats), feat_dim))), dim=-2).tolist()
        segs_feat[i] = torch.tensor(segs_feat[i], device=device)

    norm_gps_seq = pad_sequence(norm_gps_seq, batch_first=True, padding_value=0)
    trg_rid = pad_sequence(trg_rid, batch_first=True, padding_value=0).int()
    trg_onehot = pad_sequence(trg_onehot, batch_first=True, padding_value=0)
    segs_id = pad_sequence(segs_id, batch_first=True, padding_value=0).int()
    segs_feat = pad_sequence(segs_feat, batch_first=True, padding_value=0)
    segs_mask = pad_sequence(segs_mask, batch_first=True, padding_value=0)
    lengths = torch.tensor(lengths, dtype=torch.int)

    return lengths, norm_gps_seq, trg_rid, trg_onehot, segs_id, segs_feat, segs_mask