Source code for cerebras.modelzoo.data_preparation.data_preprocessing.data_dedup.to_hash

# Copyright 2022 Cerebras Systems.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import argparse
import gc
import logging
import os
import pickle
import re
import string
import sys
import threading
import time
from collections import deque
from itertools import repeat
from multiprocessing import Lock, Pool, cpu_count

import ftfy
from datasketch import MinHash
from more_itertools import chunked
from nltk import ngrams

sys.path.append(os.path.join(os.path.dirname(__file__), "../../../"))
from cerebras.modelzoo.data_preparation.nlp.hdf5_preprocessing.utils import (
    Reader,
)

dq = deque()





[docs]def custom_progress_bar(animation_delay=0.5): chars = ['|', '/', '-', '\\'] progress = 0 doc_printer_thread = threading.Thread(target=print_docs_processed) doc_printer_thread.daemon = True doc_printer_thread.start() while True: sys.stdout.write(f'\rProcessing: [{chars[progress % len(chars)]}]') sys.stdout.flush() progress += 1 time.sleep(animation_delay)
[docs]def preprocess_string(s): # Lowercase the input string s = s.lower() # Remove punctuation s = s.translate(str.maketrans("", "", string.punctuation)) # Remove consecutive spaces, newlines, tabs in the middle and in the beginning / end s = re.sub(r"\s+", " ", s.strip()) return s
[docs]def get_features(s, width): s = preprocess_string(s) return map(lambda x: "".join(x), ngrams(s, width))
[docs]def clean(s): return preprocess_string(s)
[docs]def get_documents(input_dir, jsonl_key, format, threshold, job_id, n_jobs): docs = [] gc.collect() all_files = [] for root, dirs, input_files in os.walk(input_dir): for file in input_files: all_files.append(os.path.join(root, file)) for file in all_files: parts = file.split('.')[1:] file_format = '.'.join(parts) if format == file_format: docs.append(os.path.basename(file)) no_of_files = len(docs) start = job_id * n_jobs end = start + (no_of_files // n_jobs) for index in range(start, end): input_file = docs[index] file_path = os.path.join(input_dir, input_file) tokenizable_columns = {"jsonl_key": jsonl_key} reader = Reader(file_path, tokenizable_columns) for doc_id, doc in enumerate(reader.stream_data()): if len(clean(doc)) > threshold: yield doc, file_path, doc_id
[docs]def to_minhash(chunks): gc.collect() buckets = [] documents, output_dir, width, dataset_name = chunks for doc in documents: text, file_path, doc_id = doc[0], doc[1], doc[2] file_name = file_path.split("/")[-1] output_name = f"{dataset_name}/{file_name}" text = ftfy.fix_text(text, normalization="NFC") m = MinHash(num_perm=128) m.update_batch( map(lambda x: x.encode('utf8'), get_features(text, width)) ) buckets.append( { "file_name": output_name, "doc_id": doc_id, "hash": m, } ) return buckets
[docs]def output_results(output_dir, results, chunk_id, iter): with open( f"{output_dir}/minhash_nfc/{iter}-{chunk_id}.pickle", "wb" ) as fout: pickle.dump(results, fout)
[docs]def generate_hashes(args): progress_thread = threading.Thread(target=custom_progress_bar) progress_thread.daemon = True progress_thread.start() lock = Lock() docs_processed = 0 if not os.path.exists(f"{args.output_dir}/minhash_nfc"): os.mkdir(f"{args.output_dir}/minhash_nfc") documents = get_documents( args.input_dir, args.jsonl_key, args.format, args.threshold, args.job_id, args.n_jobs, ) results = [] chunk_id = 0 gc.collect() with Pool(processes=cpu_count()) as pool: results_iterator = pool.imap( to_minhash, zip( chunked(documents, args.batch_size), repeat(args.output_dir), repeat(args.window_size), repeat(args.dataset_name), ), ) for i, chunks in enumerate(results_iterator): for chunk in chunks: if len(results) == args.batch_size: with lock: docs_processed += args.batch_size dq.append(docs_processed) output_results( args.output_dir, results, chunk_id, args.job_id ) del results gc.collect() results = [] chunk_id += 1 results.append(chunk) if results: with lock: docs_processed += len(results) logging.info(f"\nFinal document count: {docs_processed}") output_results(args.output_dir, results, chunk_id, args.job_id)
if __name__ == "__main__": logging.basicConfig(level=logging.INFO) parser = argparse.ArgumentParser() parser.add_argument( "--dataset_name", type=str, help="Name of the dataset being processed.", required=True, ) parser.add_argument( "--input_dir", type=str, help="Input directory, which contains the documents on which deduplication pipeline will be run.", required=True, ) parser.add_argument( "--output_dir", type=str, help="Output directory, where the output of the hashing process needs to be stored.", required=True, ) parser.add_argument( "--job_id", type=int, help="Job ID", default=0, required=False ) parser.add_argument( "--jsonl_key", type=str, default="text", help="JSONL key for the dataset. By default, this is set to 'text'.", required=False, ) parser.add_argument( "--format", type=str, default="jsonl", help="File format of the dataset. By default, this is set to 'jsonl'.", required=False, ) parser.add_argument( "--threshold", type=int, default=0, help="Minimum size of a document that need to be considered for deduplication. By default, this is set to 0, processing all documents.", required=False, ) parser.add_argument( "--window_size", type=int, default=6, help="Number of characters in each n-grams, used in extracting features from a document. By default, this is set to 6 (based on prior experiments).", required=False, ) parser.add_argument( "--batch_size", type=int, default=100, help="Number of documents that are processed in a batch, after which results are dumped. By default, this is set to 100.", required=False, ) parser.add_argument( "--n_jobs", type=int, default=1, help="Number of jobs to be spawned for parallel execution. By default, this is set to 1. (this is relevant only when running the deduplication pipeline across multiple machines).", required=False, ) args = parser.parse_args() generate_hashes(args)