/
Data_Container.py
187 lines (153 loc) · 7.48 KB
/
Data_Container.py
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
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader
class DataInput(object):
def __init__(self, norm_bool:bool, graph_bool:bool, norm_mode:str='minmax'):
self.norm_bool = norm_bool
self.norm_mode = norm_mode
self.norm_list = list() # initialize a list to store norm stats
self.graph_bool = graph_bool # whether the data is graph or grid based
def load_data(self, data_dir:str):
npz_data = np.load(data_dir)
print('Dataset contents:', list(npz_data.keys()))
if not self.graph_bool:
modes =[]
for mode in ['taxi', 'bike', 'scooter']:
if mode in list(npz_data.keys()):
modes.append(npz_data[mode])
mob = np.concatenate(modes, axis=-1)
else:
mob = npz_data['poi']
if not self.norm_bool:
pass
else:
mob = self.channel_wise_normalize(mob)
# store data in a dictionary
dataset = dict()
dataset['mob'] = mob
dataset['tcov'] = npz_data['meta_onehot']
dataset['mask'] = npz_data['mask']
return dataset
def normalize(self, x:np.array):
if self.norm_mode == 'minmax':
self.x_max, self.x_min = x.max(axis=0), x.min(axis=0)
x = (x - self.x_min) / (self.x_max - self.x_min)
elif self.norm_mode == 'z_score':
self.x_mean, self.x_std = x.mean(axis=0), x.std(axis=0)
x = (x - self.x_mean) / self.x_std
else:
raise ValueError
return x
def denormalize(self, x:np.array):
if self.norm_mode == 'minmax':
x = (self.x_max - self.x_min) * x + self.x_min
elif self.norm_mode == 'z_score':
x = self.x_std * x + self.x_mean
else:
raise ValueError
return x
def channel_wise_normalize(self, x:np.array):
assert len(x.shape) in [3, 4]
# x = x.swapaxes(-1, -2)
x_norm = []
for c in range(x.shape[-1]):
if self.norm_mode == 'minmax':
flow_c, c_min, c_max = self.minmax_normalize(x[...,c])
x_norm.append(flow_c)
self.norm_list.append({'min':c_min, 'max':c_max})
print(f'channel {c}, min: {c_min}, max: {c_max}')
elif self.norm_mode == 'z_score':
flow_c, c_mean, c_std = self.z_score_normalize(x[...,c])
x_norm.append(flow_c)
self.norm_list.append({'mean':c_mean, 'std':c_std})
print(f'channel {c}, mean: {c_mean}, std: {c_std}')
else:
raise ValueError
x_norm = np.stack(x_norm, axis=-1)
return x_norm # .swapaxes(-1, -2)
def channel_wise_denormalize(self, y:np.array):
# y = y.swapaxes(-1, -2)
assert y.shape[-1] == len(self.norm_list)
y_denorm = []
for c in range(y.shape[-1]):
if (self.norm_bool)&(self.norm_mode=='minmax'):
y_denorm.append(self.minmax_denormalize(y[...,c], c))
elif (self.norm_bool)&(self.norm_mode=='z_score'):
y_denorm.append(self.z_score_denormalize(y[...,c], c))
else:
raise ValueError
y_denorm = np.stack(y_denorm, axis=-1)
return y_denorm # .swapaxes(-1, -2)
def minmax_normalize(self, x:np.array): # normalize to [-1, 1]
x_max, x_min = x.max(), x.min()
x = (x - x_min) / (x_max - x_min)
#x = x * 2 - 1
return x, x_min, x_max
def minmax_denormalize(self, x:np.array, c:int):
#x = (x + 1) / 2
x = (self.norm_list[c]['max'] - self.norm_list[c]['min']) * x + self.norm_list[c]['min']
return x
def z_score_normalize(self, x:np.array): # normalize to N(0, 1)
x_mean, x_std = x.mean(), x.std()
x = (x - x_mean) / x_std
return x, x_mean, x_std
def z_score_denormalize(self, x:np.array, c:int):
x = x * self.norm_list[c]['std'] + self.norm_list[c]['mean']
return x
class MobDataset(Dataset):
def __init__(self, inputs:dict, output:torch.Tensor, mode:str, mode_len:dict):
self.mode = mode
self.mode_len = mode_len
self.inputs, self.output = self.prepare_xy(inputs, output)
def __len__(self):
return self.mode_len[self.mode]
def __getitem__(self, item):
return self.inputs['x_seq'][item], self.inputs['t_x'][item], self.inputs['t_y'][item], self.output[item]
def prepare_xy(self, inputs:dict, output:torch.Tensor):
if self.mode == 'train':
start_idx = 0
elif self.mode == 'validate':
start_idx = self.mode_len['train']
else: # test
start_idx = self.mode_len['train']+self.mode_len['validate']
x = dict()
x['x_seq'] = inputs['x_seq'][start_idx : (start_idx + self.mode_len[self.mode])]
x['t_x'] = inputs['t_x'][start_idx : (start_idx + self.mode_len[self.mode])]
x['t_y'] = inputs['t_y'][start_idx: (start_idx + self.mode_len[self.mode])]
y = output[start_idx : start_idx + self.mode_len[self.mode]]
return x, y
class DataGenerator(object):
def __init__(self, obs_len:int, pred_len, data_split_ratio:tuple):
self.obs_len = obs_len
self.pred_len = pred_len
self.data_split_ratio = data_split_ratio
def split2len(self, data_len:int):
mode_len = dict()
mode_len['validate'] = int(self.data_split_ratio[1]/sum(self.data_split_ratio) * data_len)
mode_len['test'] = int(self.data_split_ratio[2]/sum(self.data_split_ratio) * data_len)
mode_len['train'] = data_len - mode_len['validate'] - mode_len['test']
return mode_len
def get_data_loader(self, data:dict, params:dict):
feat_dict = dict()
# flows
x_seq, y_seq = self.get_feats(data['mob'].reshape(data['mob'].shape[0], params['H']*params['W'], params['C']))
feat_dict['x_seq'] = torch.from_numpy(np.asarray(x_seq)).float().to(params['device']) if params['device'].startswith('cuda') else torch.from_numpy(np.asarray(x_seq)).float()
y_seq = torch.from_numpy(np.asarray(y_seq)).float().to(params['device']) if params['device'].startswith('cuda') else torch.from_numpy(np.asarray(y_seq)).float()
# time covariates
t_x, t_y = self.get_feats(data['tcov'])
feat_dict['t_x'] = torch.from_numpy(np.asarray(t_x)).float().to(params['device']) if params['device'].startswith('cuda') else torch.from_numpy(np.asarray(t_x)).float()
feat_dict['t_y'] = torch.from_numpy(np.asarray(t_y)).float().to(params['device']) if params['device'].startswith('cuda') else torch.from_numpy(np.asarray(t_y)).float()
mode_len = self.split2len(data_len=y_seq.shape[0])
data_loader = dict() # data_loader for [train, validate, test]
for mode in ['train', 'validate', 'test']:
dataset = MobDataset(inputs=feat_dict, output=y_seq,
mode=mode, mode_len=mode_len)
data_loader[mode] = DataLoader(dataset=dataset, batch_size=params['batch_size'], shuffle=False, drop_last=True) # drop incomplete batch
# data loading default: single-processing | for multi-processing: num_workers=pos_int or pin_memory=True (GPU)
return data_loader
def get_feats(self, data:np.array):
x, y = [], []
for i in range(self.obs_len, data.shape[0]-self.pred_len):
x.append(data[i-self.obs_len : i])
y.append(data[i : i+self.pred_len])
return x, y