Datasets:
michaelmior
commited on
Finalize train/test split script
Browse files- train_split.py +66 -13
train_split.py
CHANGED
@@ -1,7 +1,10 @@
|
|
1 |
import argparse
|
|
|
|
|
2 |
import json
|
3 |
import os
|
4 |
from pathlib import Path
|
|
|
5 |
|
6 |
import numpy as np
|
7 |
import pybktree
|
@@ -17,59 +20,109 @@ def files_list():
|
|
17 |
return files
|
18 |
|
19 |
|
20 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
files = files_list()
|
22 |
|
|
|
23 |
if similarity:
|
24 |
tree = pybktree.BKTree(
|
25 |
lambda a, b: Levenshtein.distance(a, b) / max(len(a), len(b))
|
26 |
)
|
27 |
|
|
|
28 |
uf = unionfind.UnionFind()
|
29 |
-
|
|
|
|
|
|
|
|
|
30 |
for schema_file in tqdm.tqdm(files):
|
31 |
path_str = str(schema_file)
|
32 |
-
|
|
|
|
|
|
|
33 |
uf.add(str(schema_file))
|
34 |
-
if
|
35 |
-
|
|
|
36 |
else:
|
37 |
-
|
|
|
|
|
38 |
|
|
|
39 |
if similarity:
|
40 |
tree.add((str(schema_file), open(schema_file).read().strip()))
|
41 |
|
42 |
-
del
|
43 |
|
44 |
# Optionally group together similar files
|
45 |
if similarity:
|
|
|
46 |
for schema_file in tqdm.tqdm(files):
|
47 |
path_str = str(schema_file)
|
48 |
data = open(schema_file).read().strip()
|
|
|
|
|
49 |
for other_path, _ in tree.find(data, similarity):
|
50 |
uf.union(path_str, other_path)
|
51 |
|
|
|
52 |
all_schemas = list()
|
53 |
schema_groups = list()
|
54 |
for group, schemas in enumerate(uf.components()):
|
55 |
all_schemas.extend(schemas)
|
56 |
schema_groups.extend([group] * len(schemas))
|
57 |
|
|
|
58 |
all_schemas = np.array(all_schemas)
|
59 |
schema_groups = np.array(schema_groups)
|
60 |
gss = GroupShuffleSplit(n_splits=1, train_size=split, random_state=seed)
|
61 |
(train_indexes, test_indexes) = next(gss.split(all_schemas, groups=schema_groups))
|
62 |
|
63 |
-
|
64 |
-
|
65 |
-
)
|
66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
|
69 |
if __name__ == "__main__":
|
70 |
parser = argparse.ArgumentParser()
|
71 |
parser.add_argument("--similarity", default=None, type=float)
|
72 |
-
parser.add_argument("--seed", default=
|
73 |
parser.add_argument("--split", default=0.8, type=float)
|
|
|
74 |
args = parser.parse_args()
|
75 |
-
main(args.similarity, args.split, args.seed)
|
|
|
1 |
import argparse
|
2 |
+
import csv
|
3 |
+
import gzip
|
4 |
import json
|
5 |
import os
|
6 |
from pathlib import Path
|
7 |
+
import sys
|
8 |
|
9 |
import numpy as np
|
10 |
import pybktree
|
|
|
20 |
return files
|
21 |
|
22 |
|
23 |
+
def write_schemas(filename, schema_list, schema_data):
|
24 |
+
sys.stderr.write(f"Writing {filename}…\n")
|
25 |
+
with gzip.open(filename, "wt") as f:
|
26 |
+
for schema in tqdm.tqdm(list(schema_list)):
|
27 |
+
filename = str(os.path.join(*Path(schema).parts[1:]))
|
28 |
+
data = schema_data[filename]
|
29 |
+
schema = open(schema).read()
|
30 |
+
obj = {
|
31 |
+
"repository": data["repository"],
|
32 |
+
"commit": data["commit"],
|
33 |
+
"path": data["path"],
|
34 |
+
"repoStars": data["repoStars"],
|
35 |
+
"repoLastFetched": data["repoLastFetched"],
|
36 |
+
"content": schema,
|
37 |
+
}
|
38 |
+
json.dump(obj, f)
|
39 |
+
f.write("\n")
|
40 |
+
|
41 |
+
|
42 |
+
def main(similarity, split, seed, repo_file):
|
43 |
files = files_list()
|
44 |
|
45 |
+
# Prepare a BK Tree if we're doing similarity grouping
|
46 |
if similarity:
|
47 |
tree = pybktree.BKTree(
|
48 |
lambda a, b: Levenshtein.distance(a, b) / max(len(a), len(b))
|
49 |
)
|
50 |
|
51 |
+
# Initialize a union-find data structure
|
52 |
uf = unionfind.UnionFind()
|
53 |
+
|
54 |
+
# Track the first schema added to each org so we can group them
|
55 |
+
org_map = {}
|
56 |
+
|
57 |
+
sys.stderr.write("Grouping by repository…\n")
|
58 |
for schema_file in tqdm.tqdm(files):
|
59 |
path_str = str(schema_file)
|
60 |
+
|
61 |
+
# Get the organization name from the path
|
62 |
+
org = schema_file.parts[1:3]
|
63 |
+
|
64 |
uf.add(str(schema_file))
|
65 |
+
if org not in org_map:
|
66 |
+
# Track the first schema for this organization
|
67 |
+
org_map[org] = str(schema_file)
|
68 |
else:
|
69 |
+
# Merge with the previous group if this
|
70 |
+
# organization has been seen before
|
71 |
+
uf.union(org_map[org], str(schema_file))
|
72 |
|
73 |
+
# Add to the BK Tree
|
74 |
if similarity:
|
75 |
tree.add((str(schema_file), open(schema_file).read().strip()))
|
76 |
|
77 |
+
del org_map
|
78 |
|
79 |
# Optionally group together similar files
|
80 |
if similarity:
|
81 |
+
sys.stderr.write("Grouping similar files…\n")
|
82 |
for schema_file in tqdm.tqdm(files):
|
83 |
path_str = str(schema_file)
|
84 |
data = open(schema_file).read().strip()
|
85 |
+
|
86 |
+
# Find similar schemas for this schema and group them together
|
87 |
for other_path, _ in tree.find(data, similarity):
|
88 |
uf.union(path_str, other_path)
|
89 |
|
90 |
+
# Produce a list of schemas and their associated groups
|
91 |
all_schemas = list()
|
92 |
schema_groups = list()
|
93 |
for group, schemas in enumerate(uf.components()):
|
94 |
all_schemas.extend(schemas)
|
95 |
schema_groups.extend([group] * len(schemas))
|
96 |
|
97 |
+
# Split the schemas into training and test
|
98 |
all_schemas = np.array(all_schemas)
|
99 |
schema_groups = np.array(schema_groups)
|
100 |
gss = GroupShuffleSplit(n_splits=1, train_size=split, random_state=seed)
|
101 |
(train_indexes, test_indexes) = next(gss.split(all_schemas, groups=schema_groups))
|
102 |
|
103 |
+
test_schemas = all_schemas[test_indexes]
|
104 |
+
test_groups = schema_groups[test_indexes]
|
105 |
+
gss = GroupShuffleSplit(n_splits=1, train_size=0.5, random_state=seed)
|
106 |
+
(test_indexes, val_indexes) = next(gss.split(test_schemas, groups=test_groups))
|
107 |
+
|
108 |
+
schema_data = {}
|
109 |
+
with open(repo_file) as csvfile:
|
110 |
+
reader = csv.DictReader(csvfile)
|
111 |
+
for row in reader:
|
112 |
+
filename = os.path.join(row["repository"], row["path"])
|
113 |
+
schema_data[filename] = row
|
114 |
+
|
115 |
+
# Write the train and test sets
|
116 |
+
write_schemas("train.jsonl.gz", all_schemas[train_indexes], schema_data)
|
117 |
+
write_schemas("test.jsonl.gz", test_schemas[test_indexes], schema_data)
|
118 |
+
write_schemas("validation.jsonl.gz", test_schemas[val_indexes], schema_data)
|
119 |
|
120 |
|
121 |
if __name__ == "__main__":
|
122 |
parser = argparse.ArgumentParser()
|
123 |
parser.add_argument("--similarity", default=None, type=float)
|
124 |
+
parser.add_argument("--seed", default=94, type=int)
|
125 |
parser.add_argument("--split", default=0.8, type=float)
|
126 |
+
parser.add_argument("--repo_file", default="repos.csv")
|
127 |
args = parser.parse_args()
|
128 |
+
main(args.similarity, args.split, args.seed, args.repo_file)
|