-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset.py
173 lines (148 loc) · 6.1 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
167
168
169
170
171
172
173
import torch
import pandas as pd
import random
import os
from PIL import Image
import numpy as np
from abc import ABC, abstractmethod
import numpy as np
class MimicCxrDataset(ABC, torch.utils.data.Dataset):
def __init__(self, path: str, split: str = "train"):
self.path = path
self.error = True
self.file_path = os.path.join(self.path, "files")
self.dataframe = pd.read_csv(os.path.join(self.path, "random_sampled.csv"), index_col=0).sort_index()
self.dataframe = self.dataframe[self.dataframe["split"] == split]
# leggo le label assegnate ai vari record
self.labels_dataframe = pd.read_csv(os.path.join(self.path, "mimic-cxr-2.0.0-negbio.csv"))
self.labels_dataframe = self.labels_dataframe.loc[
self.labels_dataframe["study_id"].isin(self.dataframe["study_id"])
]
def _get_text(self, idx) -> str:
self.__check_valid_index(idx)
text_row = self.dataframe.iloc[idx]
txt_path = os.path.join(
self.file_path,
f"p{str(text_row.subject_id)[:2]}",
f"p{text_row.subject_id}",
f"s{text_row.study_id}.txt",
)
with open(txt_path, "r") as f:
text = f.read()
return text
def _get_image(self, idx) -> np.array:
self.__check_valid_index(idx)
image_row = self.dataframe.iloc[idx]
image_path = os.path.join(
self.file_path,
f"p{str(image_row.subject_id)[:2]}",
f"p{image_row.subject_id}",
f"s{image_row.study_id}",
f"{image_row.dicom_id}.jpg",
)
image = Image.open(image_path)
image = np.array(image)
image = np.stack((image,) * 3, axis=-1)
return image
def get_labels(self, idx) -> np.array:
study_id = self.get_study_id(idx)
return self.labels_dataframe.loc[self.labels_dataframe.study_id == study_id].iloc[0, 2:].values
def get_study_id(self, idx) -> int:
return self.dataframe.iloc[idx].study_id
def __check_valid_index(self, idx) -> None:
assert idx < len(self.dataframe), f"Index out of range ({idx} > {len(self.dataframe)-1})"
@abstractmethod
def __getitem__(self, index):
pass
@abstractmethod
def __len__(self):
pass
class MimicCxrPretrainingDataset(MimicCxrDataset):
def __getitem__(self, idx):
texts = []
images = []
labels = []
if isinstance(idx, torch.Tensor):
idx = idx.tolist()
elif isinstance(idx, int):
idx = [idx]
else:
print("Error, type not expected: " + type(idx))
for i in idx:
if i >= len(self.dataframe):
text_index = i % len(self.dataframe)
image_index = self._get_different_image_index(text_index)
else:
text_index = i
image_index = i
# text_row = self.dataframe.iloc[i]
# image_row = self.dataframe.iloc[i]
texts.append(self._get_text(text_index))
images.append(self._get_image(image_index))
labels.append(int(i < len(self.dataframe)))
return {"texts": texts, "images": images, "next_sentence_labels": labels}
def _get_different_image_index(self, text_index) -> int:
text_row = self.dataframe.iloc[text_index]
labels_to_check = (
self.labels_dataframe.loc[self.labels_dataframe["study_id"] == text_row.study_id] == 1.0
)
searching = True
while searching:
image_index = random.randint(0, len(self.dataframe) - 1)
image_row = self.dataframe.iloc[image_index]
image_labels = self.labels_dataframe.loc[self.labels_dataframe["study_id"] == image_row.study_id]
searching = (
len(labels_to_check.index) > 0
and (image_labels[labels_to_check].squeeze() == 1.0).all(axis=None)
) or text_row.study_id == image_row.study_id
if searching:
self.error = False
return image_index
def __len__(self):
return 2 * len(self.dataframe)
def checkerror(self):
return self.error
class MimicCxrPretrainingDatasetAnyLabels(MimicCxrPretrainingDataset):
def _get_different_image_index(self, text_index) -> int:
text_row = self.dataframe.iloc[text_index]
labels_to_check = (
self.labels_dataframe.loc[self.labels_dataframe["study_id"] == text_row.study_id] == 1.0
)
searching = True
while searching:
image_index = random.randint(0, len(self.dataframe) - 1)
image_row = self.dataframe.iloc[image_index]
image_labels = self.labels_dataframe.loc[self.labels_dataframe["study_id"] == image_row.study_id]
searching = (
len(labels_to_check.index) > 0
and (image_labels[labels_to_check].squeeze() == 1.0).any(axis=None)
) or text_row.study_id == image_row.study_id
if searching:
self.error = False
return image_index
class MimicCxrPretrainingDatasetRandom(MimicCxrPretrainingDataset):
def _get_different_image_index(self, text_index) -> int:
text_row = self.dataframe.iloc[text_index]
searching = True
while searching:
image_index = random.randint(0, len(self.dataframe) - 1)
image_row = self.dataframe.iloc[image_index]
searching = text_row.study_id == image_row.study_id
return image_index
class MimicCxrMetricLearningDataset(MimicCxrDataset):
def __getitem__(self, idx):
text = self._get_text(idx)
image = self._get_image(idx)
study_id = self.get_study_id(idx)
return image, text, study_id
def __len__(self):
return len(self.dataframe)
class MimicCxrDatasetBasic(MimicCxrDataset):
def __get_label(self, idx) -> int:
return int(self.dataframe.iloc[idx].study_id)
def __getitem__(self, idx):
text = self._get_text(idx)
image = self._get_image(idx)
return image, text, idx
def __len__(self):
return len(self.dataframe)