-
Notifications
You must be signed in to change notification settings - Fork 14
/
Copy pathconvert_to_meds.py
162 lines (152 loc) · 6.27 KB
/
convert_to_meds.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
"""
Convert EHRSHOT to MEDS format
Note: Takes ~10 mins
"""
import datetime
from tqdm import tqdm
import json
import jsonschema
import femr.datasets
from typing import List, Dict, Optional, Any, Set
import pandas as pd
import pyarrow as pa
import pyarrow.parquet as pq
import os
from meds import (
code_metadata_schema,
data_schema,
dataset_metadata_schema,
subject_split_schema,
label_schema,
train_split,
tuning_split,
held_out_split,
)
path_to_output_dir: str = '../../EHRSHOT_ASSETS/meds'
os.makedirs(path_to_output_dir, exist_ok=True)
os.makedirs(os.path.join(path_to_output_dir, 'metadata'), exist_ok=True)
os.makedirs(os.path.join(path_to_output_dir, 'data'), exist_ok=True)
print(f"Saving output to {path_to_output_dir}")
# Dataset
metadata: Dict[str, str] = {
"dataset_name": "EHRSHOT",
"dataset_version": "1.0",
"etl_name": "EHRSHOT => MEDS ETL",
"etl_version": "1.0",
"meds_version" : "0.3.3",
"created_at" : str(datetime.datetime.now()),
}
jsonschema.validate(instance=metadata, schema=dataset_metadata_schema)
print("Writing dataset.json")
json.dump(metadata, open(os.path.join(path_to_output_dir, 'metadata/dataset.json'), 'w'))
# Splits
print("Converting splits...")
df_splits = pd.read_csv('../../EHRSHOT_ASSETS/splits_ehrshot/person_id_map.csv')
df_splits = df_splits.rename(columns={ 'omop_person_id' : 'subject_id', })
df_splits['split'] = df_splits['split'].apply(lambda x: train_split if x == 'train' else (tuning_split if x == 'val' else held_out_split))
pa_splits = pa.Table.from_pandas(df_splits, schema=subject_split_schema)
print("Writing metadata/subject_splits.parquet")
pq.write_table(pa_splits, os.path.join(path_to_output_dir, 'metadata/subject_splits.parquet'))
## Sanity check
df_ = pd.read_parquet(os.path.join(path_to_output_dir, 'metadata/subject_splits.parquet'))
assert pa_splits.schema.equals(subject_split_schema), "subject split schema does not match"
assert df_splits.shape[0] == df_.shape[0], "splits shape does not match"
print(df_.head())
print("Shape:", df_.shape)
# Events + Codes
print("Converting events...")
femr_db = femr.datasets.PatientDatabase('../../EHRSHOT_ASSETS/femr/extract')
events: List[Dict] = []
codes: List[Dict] = []
codes_set: Set[str] = set()
for pid in tqdm(femr_db, total=len(femr_db), desc='events'):
for event in femr_db[pid].events:
## Event
e: Dict[str, Any] = {
'subject_id' : pid,
'time' : event.start,
'code' : event.code,
'unit' : event.unit,
'omop_table' : event.omop_table,
'numeric_value' : None,
}
if event.value is not None:
if isinstance(event.value, str):
e['text_value'] = event.value
elif isinstance(event.value, float) or isinstance(event.value, int):
e['numeric_value'] = event.value
events.append(e)
## Code
if event.code not in codes_set:
codes.append({
'code' : event.code,
'description' : '',
})
codes_set.add(event.code)
## Events
print("Writing data/data.parquet")
events_schema = data_schema([
("text_value", pa.string()),
("unit", pa.string()),
("omop_table", pa.string())
])
pa_events = pa.Table.from_pylist(events, schema=events_schema)
pq.write_table(pa_events, os.path.join(path_to_output_dir, 'data/data.parquet'))
## Sanity check
df_ = pd.read_parquet(os.path.join(path_to_output_dir, 'data/data.parquet'))
assert pa_events.schema.equals(events_schema), "events schema does not match"
assert pa_events.shape[0] == df_.shape[0], "events shape does not match"
print(df_.head())
print("Shape:", df_.shape)
## Codes
print("Writing metadata/codes.parquet")
codes_schema = code_metadata_schema()
pa_codes = pa.Table.from_pylist(codes, schema=codes_schema)
pq.write_table(pa_codes, os.path.join(path_to_output_dir, 'metadata/codes.parquet'))
## Sanity check
df_ = pd.read_parquet(os.path.join(path_to_output_dir, 'metadata/codes.parquet'))
assert pa_codes.schema.equals(codes_schema), "codes schema does not match"
assert pa_codes.shape[0] == df_.shape[0], "events shape does not match"
print(df_.head())
print("Shape:", df_.shape)
# Labels
print("Converting labels...")
path_to_labels_dir: str = '../../EHRSHOT_ASSETS/benchmark_ehrshot/'
for task in tqdm(os.listdir(path_to_labels_dir), desc='labels'):
if not os.path.isdir(os.path.join(path_to_labels_dir, task)):
continue
path_to_labels: str = os.path.join(path_to_labels_dir, task, 'labeled_patients.csv')
df_labels = pd.read_csv(path_to_labels)
df_labels = df_labels.rename(columns={ 'patient_id' : 'subject_id', })
label_type: str = df_labels['label_type'].iloc[0]
if label_type == 'boolean':
df_labels = df_labels.rename(columns={ 'value' : 'boolean_value', })
df_labels['integer_value'] = None
df_labels['float_value'] = None
df_labels['categorical_value'] = None
elif label_type == 'categorical':
df_labels = df_labels.rename(columns={ 'value' : 'integer_value', })
df_labels['boolean_value'] = None
df_labels['float_value'] = None
df_labels['categorical_value'] = None
else:
raise ValueError(f"Unexpected label type: {label_type}")
df_labels = df_labels.drop(columns=['label_type'])
df_labels['prediction_time'] = pd.to_datetime(df_labels['prediction_time'])
pa_labels = pa.Table.from_pandas(df_labels, schema=label_schema)
print(f"Writing labels/{task}/labels.parquet")
os.makedirs(os.path.join(path_to_output_dir, 'labels', task), exist_ok=True)
pq.write_table(pa_labels, os.path.join(path_to_output_dir, 'labels', task, 'labels.parquet'))
## Sanity check
df_ = pd.read_parquet(os.path.join(path_to_output_dir, 'labels', task, 'labels.parquet'))
assert pa_labels.schema.equals(label_schema), "labels schema does not match"
assert df_labels.shape[0] == df_.shape[0], "labels shape does not match"
print(df_.head())
print("Shape:", df_.shape)
# Confirm it worked
import meds_reader
print("Running meds_reader to confirm MEDS extract is valid...")
status: int = os.system(f"meds_reader_convert {path_to_output_dir} {path_to_output_dir}_db --num_threads 5")
if status != 1:
raise ValueError("Failed to convert to MEDS DB -- Couldn't run meds_reader on it")
print("Done!")