-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset.py
166 lines (127 loc) · 6.61 KB
/
dataset.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
import torch
import numpy as np
from torch.utils.data import Dataset
from scipy.sparse import coo_matrix
from IPython import embed
def coo2tensor(coo, device):
i = torch.LongTensor(np.vstack((coo.row, coo.col)))
v = torch.FloatTensor(coo.data)
shape = coo.shape
return torch.sparse_coo_tensor(i, v, torch.Size(shape), device=device)
def csr2tensor(csr, device):
return coo2tensor(csr.tocoo(), device)
def csc2tensor(csc, device):
return coo2tensor(csc.tocoo(), device)
def normalize(data, vocab_cum, bow_norm=True):
if not bow_norm:
return data
normalized_data = torch.zeros(data.shape, device=data.device)
for i, _ in enumerate(vocab_cum[:-1]):
data_i = data[:,vocab_cum[i]:vocab_cum[i+1]]
sum_i = data_i.sum(-1).unsqueeze(-1) + 1e-6
normalized_data[:,vocab_cum[i]:vocab_cum[i+1]] = data_i / sum_i
return normalized_data
class NontemporalDataset(Dataset):
def __init__(self, phase, npy_file, code_type_info, device, transform=None, drug_imputation=False, drug_count_thr=1):
self.phase = phase
self.device = device
if phase == 'test':
npy_file, npy_file_1, npy_file_2 = npy_file
self.bow = np.load(npy_file, allow_pickle=True).item()
self.bow_1 = np.load(npy_file_1, allow_pickle=True).item()
self.bow_2 = np.load(npy_file_2, allow_pickle=True).item()
# FIXME: 10-14 sparse tensor dataset
self.bow_1 = csr2tensor(self.bow_1, device)
self.bow_2 = csr2tensor(self.bow_2, device)
else:
self.bow = np.load(npy_file, allow_pickle=True).item()
self.code_types, self.vocab_size, self.vocab_cum = code_type_info
self.N = self.bow.shape[0]
self.V = self.bow.shape[1]
self.transform = transform
if phase == 'train':
bow_coo = self.bow.tocoo()
self.stack_bow = {}
for i, c in enumerate(self.code_types):
chs = (bow_coo.col >= self.vocab_cum[i]) * (bow_coo.col < self.vocab_cum[i+1])
stack_row_i = list(bow_coo.row[chs])
stack_col_i = list(bow_coo.col[chs] - self.vocab_cum[i])
stack_data_i = list(bow_coo.data[chs])
self.stack_bow[c] = coo_matrix((stack_data_i, (stack_row_i, stack_col_i)), shape=(self.N, self.vocab_size[i]))
if drug_imputation:
self.drug_count_thr = drug_count_thr
non_drug_count = np.array(self.bow[:,:self.vocab_cum[1]].sum(1)).squeeze(1)
drug_count = np.array(self.bow[:,self.vocab_cum[1]:].astype('bool').sum(1)).squeeze(1)
# np.save(open('train_drug_count.npy','wb'), drug_count[non_drug_count>0])
drug_count = drug_count * (drug_count >= self.drug_count_thr)
self.imputation_idx = np.nonzero(non_drug_count * drug_count)[0]
self.imputation_data = self.bow[self.imputation_idx]
imputation_drug_occur = self.imputation_data.astype('bool')[:, self.vocab_cum[1]:]
self.imputation_drug_freq = torch.from_numpy(np.array(imputation_drug_occur.tocsc().sum(0)).squeeze(0)).to(device)
print(f"# (imputataion train samples (drug_counts>={drug_count_thr})): {self.imputation_data.shape[0]}")
self.imputation_data = csr2tensor(self.imputation_data, device).coalesce()
if phase == 'test':
if drug_imputation:
self.drug_count_thr = drug_count_thr
non_drug_count = np.array(self.bow[:,:self.vocab_cum[1]].sum(1)).squeeze(1)
drug_count = np.array(self.bow[:,self.vocab_cum[1]:].astype('bool').sum(1)).squeeze(1)
# drug_count = np.array(self.bow[:,self.vocab_cum[1]:].sum(1)).squeeze(1)
# np.save(open('test_drug_count.npy','wb'), drug_count[non_drug_count>0])
drug_count = drug_count * (drug_count >= drug_count_thr)
self.imputation_idx = np.nonzero(non_drug_count * drug_count)[0]
self.imputation_data = self.bow[self.imputation_idx]
imputation_drug_occur = self.imputation_data.astype('bool')[:, self.vocab_cum[1]:]
self.imputation_drug_freq = torch.from_numpy(np.array(imputation_drug_occur.tocsc().sum(0)).squeeze(0)).to(device)
print(f"# (imputataion test samples (drug_counts>={drug_count_thr})): {self.imputation_data.shape[0]}")
self.imputation_data = csr2tensor(self.imputation_data, device).coalesce()
self.bow = csr2tensor(self.bow, device).coalesce()
def __len__(self):
return self.N
def __getitem__(self, idx):
# data = torch.from_numpy(self.bow[idx].todense()).squeeze(0)
# FIXME: 10-14 sparse tensor dataset
# data = csr2tensor(self.bow[idx])
data = self.bow[idx]
sample = {'Data': data}
if self.phase == 'test':
# FIXME: 10-14 sparse tensor dataset
# data_1 = csr2tensor(self.bow_1[idx])
# data_2 = csr2tensor(self.bow_2[idx])
data_1 = self.bow_1[idx]
data_2 = self.bow_2[idx]
sample.update({'Data_1': data_1, 'Data_2': data_2})
if self.transform:
sample = self.transform(sample)
return sample, idx
class TemporalDataset(Dataset):
def __init__(self, phase, npy_file, code_type_info, transform=None):
self.aggregate = aggregate
self.bow = np.load(npy_file, allow_pickle=True)
self.code_types, self.vocab_size, self.vocab_cum = code_type_info
self.transform = transform
self.phase = phase
# FIXME: copied from non-temporal case
T = 11
N = self.bow.shape[0]
V = self.bow.shape[1]
self.stack_bow = {}
for i, c in enumerate(self.code_types):
chs = (self.bow.col >= self.vocab_cum[i]) * (self.bow.col < self.vocab_cum[i+1])
stack_row_i = list(self.bow.row[chs])
stack_col_i = list(self.bow.col[chs] - self.vocab_cum[i])
stack_data_i += list(doc.data[chs])
self.stack_bow[c] = coo_matrix((stack_data_i, (stack_row_i, stack_col_i)), shape=(N, self.vocab_size[i]))
def __len__(self):
return len(self.bow)
def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
data = self.bow[idx]
if isinstance(data, coo_matrix):
data = coo2tensor(data)
else:
data = torch.from_numpy(data)
sample = {'Data': data}
if self.transform:
sample = self.transform(sample)
return sample, idx