-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathCustomDataset.py
163 lines (129 loc) · 6.44 KB
/
CustomDataset.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
import os
import pandas as pd
import torch
from torch.utils.data import Dataset, Subset
import random
random.seed(1)
class GraphDataset(Dataset):
def __init__(self, root, model, leave_out=[]):
self.root = root
self.model = model
domains = [
'barman',
'blocksworld',
'childsnack',
'data_network',
'depots',
'driverlog',
'floortile',
'gripper',
'hiking',
'logistics',
'maintenance',
'miconic',
'openstacks',
'parking',
'rovers',
'scanalyzer',
'transport',
'visitall',
'woodworking',
'zenotravel'
]
idx = 0
self.datapoints = pd.DataFrame(columns=['domain', 'file', 'grounded_size'])
if self.model == 'lifted':
for domain in domains:
for file in os.listdir(os.path.join(self.root, 'labeled_lifted_data', domain)):
self.datapoints.loc[idx] = [domain, file]
idx += 1
elif self.model == 'grounded':
for domain in domains:
for file in os.listdir(os.path.join(self.root, 'labeled_grounded_data', domain)):
grounded_size = torch.load(os.path.join(self.root, 'labeled_grounded_data', domain, file)).x.shape[0]
self.datapoints.loc[idx] = [domain, file, grounded_size]
idx += 1
else:
for domain in domains:
lifted_files = set(os.listdir(os.path.join(self.root, 'labeled_lifted_data', domain)))
grounded_files = set(os.listdir(os.path.join(self.root, 'labeled_grounded_data', domain)))
union = list(lifted_files.intersection(grounded_files))
for file in union:
grounded_size = torch.load(os.path.join(self.root, 'labeled_grounded_data', domain, file)).x.shape[0]
self.datapoints.loc[idx] = [domain, file, grounded_size]
idx += 1
def __len__(self):
return len(self.datapoints)
def __getitem__(self, index):
if self.model == 'dual':
domain = self.datapoints.loc[index].domain
file = self.datapoints.loc[index].file
lifted_path = os.path.join(self.root, 'labeled_lifted_data', domain, self.datapoints.loc[index].file)
grounded_path = os.path.join(self.root, 'labeled_grounded_data', domain, self.datapoints.loc[index].file)
lifted_data = torch.load(lifted_path)
grounded_data = torch.load(grounded_path)
return lifted_data, grounded_data, domain, file
elif self.model == 'lifted':
domain = self.datapoints.loc[index].domain
file = self.datapoints.loc[index].file
path = os.path.join(self.root, 'labeled_lifted_data', domain, self.datapoints.loc[index].file)
lifted_data = torch.load(path)
return lifted_data, lifted_data, domain, file
else:
domain = self.datapoints.loc[index].domain
file = self.datapoints.loc[index].file
path = os.path.join(self.root, 'labeled_grounded_data', domain, self.datapoints.loc[index].file)
grounded_data = torch.load(path)
return grounded_data, grounded_data, domain, file
def alternative_split(self, ratios:tuple, shuffle=False):
train, val, test = ratios
train_set = []
val_set = []
test_set = []
domains = self.datapoints.domain.unique()
for domain in domains:
subset_indices = self.datapoints.index[self.datapoints.domain == domain].to_list()
if shuffle:
random.shuffle(subset_indices)
domain_length = len(subset_indices)
train_set.append(subset_indices[:int(train*domain_length)])
val_set.append(subset_indices[int(train*domain_length):int(train*domain_length)+int(val*domain_length)])
test_set.append(subset_indices[int(train*domain_length)+int(val*domain_length):])
train_set = [item for items in train_set for item in items]
test_set = [item for items in test_set for item in items]
val_set = [item for items in val_set for item in items]
return Subset(self, train_set), Subset(self, val_set), Subset(self, test_set)
def graph_size_split(self, ratios:tuple):
"""
Makes split according to the grounded graph size (assumed larger graphs are more complex).
"""
train, val, test = ratios
train_set = []
val_set = []
test_set = []
domains = self.datapoints.domain.unique()
sorted_set = self.datapoints.sort_values(by='grounded_size', ascending=True)
for domain in domains:
subset_indices = sorted_set.index[sorted_set.domain == domain].to_list()
subset_domains = sorted_set[sorted_set.domain == domain].domain.to_list()
subset_files = sorted_set[sorted_set.domain == domain].file.to_list()
domain_length = len(subset_indices)
train_set.append(subset_indices[:int(train*domain_length)])
val_set.append(subset_indices[int(train*domain_length):int(train*domain_length)+int(val*domain_length)])
test_set.append(subset_indices[int(train*domain_length)+int(val*domain_length):])
train_set = [item for items in train_set for item in items]
test_set = [item for items in test_set for item in items]
val_set = [item for items in val_set for item in items]
return Subset(self, train_set), Subset(self, val_set), Subset(self, test_set)
def transfer_split(self, domain='driverlog', graph_size=True):
domains = self.datapoints.domain.unique()
domains = [i for i in domains if i != domain]
data = self.datapoints
if graph_size:
data = self.datapoints.sort_values(by='grounded_size', ascending=True)
train_set = data.index[data.domain != domain].to_list()
transfer_set = data.index[data.domain == domain].to_list()
domain_length = len(transfer_set)
transfer_train_set = transfer_set[:int(0.85*domain_length)]
transfer_test_set = transfer_set[int(0.85*domain_length):]
return Subset(self, train_set), Subset(self, transfer_train_set), Subset(self, transfer_test_set)