File size: 2,913 Bytes
877a3f4 |
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 |
# Author: Roman Janík
# Script for creating splits of a CoNLL format of my dataset.
#
import argparse
import os
from random import shuffle
from itertools import islice
def get_args():
parser = argparse.ArgumentParser(description="Script for creating splits of a CoNLL format of my dataset.")
parser.add_argument("-s", "--source_file", required=True, help="Path to source Label Studio json annotations file.")
parser.add_argument("-o", "--output_dir", required=True, help="Output dir for CoNLL dataset splits.")
parser.add_argument("-m", "--shuffle", default=True, dest="shuffle", action="store_true",
help="Shuffle annotations.")
parser.add_argument("--shuffle_chunk", type=int, default=1, dest="shuffle_chunk",
help="Shuffle annotations chunk. Chunk annotations, shuffle and join back.")
parser.add_argument("-p", "--split_ratios", required=True,
help="Ratios for train/dev/test splits, nums separated by ':'")
args = parser.parse_args()
return args
def parse_splits(split_str: str):
split_list = split_str.split(":")
assert len(split_list) == 3
splits = {k: float(v) for k, v in zip(["train", "dev", "test"], split_list)}
assert sum(splits.values()) == 1.0
return splits
def batched_func(iterable, chunk_size):
iterator = iter(iterable)
return list(iter(
lambda: list(islice(iterator, chunk_size)),
list()
))
def main():
args = get_args()
print("Script for creating a CoNLL format of my dataset from text files and Label Studio annotations. "
"Script goes through page text files and their annotations json record. "
"Output CoNLL dataset split files are saved to output directory.")
splits_ratios = parse_splits(args.split_ratios)
with open(args.source_file, encoding="utf-8") as f:
content = f.read()
annotations = content.split("\n\n")
print(f"Total training examples: {len(annotations)}\nStarting documents processing...")
if args.shuffle:
if args.shuffle_chunk > 1:
chunked_annotations = batched_func(annotations, args.shuffle_chunk)
shuffle(chunked_annotations)
annotations = [item for sublist in chunked_annotations for item in sublist]
else:
shuffle(annotations)
splits = {}
lower_bound = 0
for k, ratio in splits_ratios.items():
upper_bound = lower_bound + int(ratio * len(annotations))
splits[k] = annotations[lower_bound:upper_bound]
lower_bound = upper_bound
for name, split in splits.items():
with open(os.path.join(args.output_dir, "poner_" + name + ".conll"), "w", encoding="utf-8") as f:
for example in split:
f.write(example + "\n\n")
print(f"Split '{name}' is processed.")
if __name__ == '__main__':
main()
|