p4dataset / res /minhash_deduplication.patch
Mihai-Valentin DUMITRU
initial commit
065fa03
Raw History Blame Contribute Delete
2.43 kB
diff --git a/near_deduplication/minhash_deduplication.py b/near_deduplication/minhash_deduplication.py
index e423541..e442eed 100644
--- a/near_deduplication/minhash_deduplication.py
+++ b/near_deduplication/minhash_deduplication.py
@@ -4,6 +4,7 @@
# created : 10/4/22
from __future__ import annotations
+import json
import gc
import hashlib
import logging
@@ -43,6 +44,19 @@ logger.setLevel(logging.INFO)
datasets.logging.set_verbosity_error()
+def fix_json(injsfile, outjsfile):
+ with open(injsfile, "rt") as fin:
+ lines = fin.readlines()
+
+ fixed_lines = []
+ for line in lines:
+ crt = json.dumps(json.loads(line)) + "\n"
+ fixed_lines.append(crt)
+
+ with open(outjsfile, "wt") as fout:
+ fout.writelines(fixed_lines)
+
+
def ngrams(sequence: List[str], n: int, min_ngram_size: int) -> Iterable:
"""
Directly taken from nltk package to avoid dependency.
@@ -222,6 +236,7 @@ if __name__ == "__main__":
threshold: float = typer.Option(0.7, help="Minhash threshold"),
min_ngram_size: int = typer.Option(5, help="Shorter documents will be removed"),
output: str = typer.Option(None, help="Store the deduplicated dataset"),
+ dedupjson: str = typer.Option(None, help="Store the deduplicated dataset as json"),
):
global uf
OUTPUT_BASE = Path(output or "output")
@@ -238,16 +253,7 @@ if __name__ == "__main__":
HASH_TABLES = [defaultdict(set) for _ in range(B)]
time_measures["load_dataset"] = time.time()
- ds = load_dataset(
- dataset,
- config,
- data_dir=data_dir,
- split=split,
- use_auth_token=True,
- cache_dir=cache_dir,
- revision=revision,
- num_proc=os.cpu_count(),
- )
+ ds = load_dataset("json", data_files=dataset)["train"]
time_measures["load_dataset"] = time.time() - time_measures["load_dataset"]
DATA_SIZE = len(ds)
PERMUTATIONS = np.array(
@@ -322,7 +328,8 @@ if __name__ == "__main__":
time_measures["save"] = time.time()
final_data = final_data.remove_columns(["__cluster__"])
- final_data.save_to_disk(output)
+ final_data.to_json(output)
+ fix_json(output, dedupjson)
time_measures["save"] = time.time() - time_measures["save"]
FINAL_DATA_SIZE = len(final_data)