-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathkey_datamodule.py
202 lines (170 loc) · 7.75 KB
/
key_datamodule.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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
import os
import torch
import random
import pandas as pd
from sklearn.model_selection import train_test_split
import lightning.pytorch as pl
from dataset import BigramDataset, BigramDatasetVal, BigramPlusDataset
class KeyDataModule(pl.LightningDataModule):
"""
DataModule for the Key Prediction task.
"""
def __init__(self,
root_dir: str = "",
user_cnt: int = 100,
max_seq_len: int = 50,
test_size: int = 5,
replace_prob: float = 0.1,
user_prob: float = 0.5,
batch_size: int = 512,
val_batch_size: int = 16,
val_user_cnt: int = 1000,
num_workers: int = 4,
dataset_multiplier: int = 1):
super().__init__()
self.root_dir = root_dir
self.user_cnt = user_cnt
self.max_seq_len = max_seq_len
self.test_size = test_size
self.replace_prob = replace_prob
self.user_prob = user_prob
self.batch_size = batch_size
self.val_batch_size = val_batch_size
self.val_user_cnt = val_user_cnt
self.num_workers = num_workers
self.dataset_multiplier = dataset_multiplier
self.raw_data_dir = os.path.join(self.root_dir, "clean")
self.train_dir = os.path.join(self.root_dir, "train")
self.test_dir = os.path.join(self.root_dir, "test")
def split_data(self, user_mapping):
"""
Split the data into train and test folders.
"""
train_files = os.listdir(self.train_dir)
test_files = os.listdir(self.test_dir)
for file in train_files:
os.remove(os.path.join(self.train_dir, file))
for file in test_files:
os.remove(os.path.join(self.test_dir, file))
for user_file, user_id in user_mapping.items():
user_df = pd.read_csv(os.path.join(self.raw_data_dir, user_file))
# split data into train and test
train_df, test_df = train_test_split(user_df,
test_size=self.test_size,
random_state=42)
# save train and test data
train_df.to_csv(os.path.join(self.train_dir,
str(user_id) + ".csv"),
index=False)
test_df.to_csv(os.path.join(self.test_dir,
str(user_id) + ".csv"),
index=False)
def collate_fn(self, batch):
b0, b1, feat, user, target, user_target = zip(*batch)
b0 = [torch.tensor(b) for b in b0]
b1 = [torch.tensor(b) for b in b1]
feat = [torch.tensor(f) for f in feat]
b0_padded = torch.nn.utils.rnn.pad_sequence(b0,
batch_first=True,
padding_value=0)
b1_padded = torch.nn.utils.rnn.pad_sequence(b1,
batch_first=True,
padding_value=0)
feat_padded = torch.nn.utils.rnn.pad_sequence(feat,
batch_first=True,
padding_value=0)
target_padded = torch.nn.utils.rnn.pad_sequence(target,
batch_first=True,
padding_value=-1)
target_padded = target_padded.long()
user = torch.tensor(user)
user_target = torch.tensor(user_target)
# attention mask
attn_mask = torch.nn.Transformer.generate_square_subsequent_mask(
target_padded.size(1) + 1)
# zero out the first row (= user)
attn_mask[0, :] = 0
mask = target_padded == -1
# add user mask to the batch
mask = torch.cat((torch.zeros((mask.shape[0], 1)), mask), dim=1)
return b0_padded, b1_padded, feat_padded, mask, user, attn_mask, user_target, target_padded
def collate_val_fn(self, batch):
b0s, b1s, feats, users, targets, user_target = zip(*batch)
b0 = [torch.tensor(b) for b0 in b0s for b in b0]
b1 = [torch.tensor(b) for b1 in b1s for b in b1]
feat = [torch.tensor(f) for fs in feats for f in fs]
target = [t for ts in targets for t in ts]
user = [u for us in users for u in us]
user_target = [ut for uts in user_target for ut in uts]
b0_padded = torch.nn.utils.rnn.pad_sequence(b0,
batch_first=True,
padding_value=0)
b1_padded = torch.nn.utils.rnn.pad_sequence(b1,
batch_first=True,
padding_value=0)
feat_padded = torch.nn.utils.rnn.pad_sequence(feat,
batch_first=True,
padding_value=0)
target_padded = torch.nn.utils.rnn.pad_sequence(target,
batch_first=True,
padding_value=-1)
target_padded = target_padded.long()
user = torch.tensor(user)
user_target = torch.tensor(user_target)
# attention mask
attn_mask = torch.nn.Transformer.generate_square_subsequent_mask(
target_padded.size(1) + 1)
# zero out the first row (= user)
attn_mask[0, :] = 0
# padding mask
mask = target_padded == -1
# add user mask to the batch
mask = torch.cat((torch.zeros((mask.shape[0], 1)), mask), dim=1)
return b0_padded, b1_padded, feat_padded, mask, user, attn_mask, user_target, target_padded
def prepare_data(self):
"""
Sample user_cnt users from the raw data.
Split the data from each user into train and test.
"""
if len(os.listdir(self.train_dir)) > 0:
return
else:
valid_user_files = os.listdir(self.raw_data_dir)
chosen_users = random.sample(valid_user_files, self.user_cnt)
user_mapping = {user: i for i, user in enumerate(chosen_users)}
self.split_data(user_mapping)
print("Data prepared.")
def setup(self, stage=None):
"""
Load the data from the train and test folders.
"""
if stage == "fit" or stage is None:
self.train_dataset = BigramPlusDataset(self.train_dir,
self.max_seq_len,
self.replace_prob,
self.dataset_multiplier)
self.val_dataset = BigramDatasetVal(self.test_dir,
self.max_seq_len,
self.val_user_cnt)
def train_dataloader(self):
return torch.utils.data.DataLoader(self.train_dataset,
batch_size=self.batch_size,
shuffle=True,
collate_fn=self.collate_fn,
num_workers=self.num_workers,
persistent_workers=True,
pin_memory=True)
def val_dataloader(self):
assert self.val_batch_size == 1, "val_batch_size must be 1"
return torch.utils.data.DataLoader(self.val_dataset,
batch_size=self.val_batch_size,
shuffle=False,
collate_fn=self.collate_val_fn,
num_workers=self.num_workers,
pin_memory=True)
def test_dataloader(self):
return torch.utils.data.DataLoader(self.val_dataset,
batch_size=self.val_batch_size,
shuffle=False,
collate_fn=self.collate_val_fn,
num_workers=self.num_workers)