107 lines
3.3 KiB
Python
107 lines
3.3 KiB
Python
import argparse
|
|
import random
|
|
import time
|
|
import unicodedata
|
|
import sys
|
|
|
|
import numpy as np
|
|
|
|
|
|
SHINGLE_SIZE = 5 # Known as k
|
|
PERMUTATIONS_COUNT = 3
|
|
|
|
|
|
def parse_args(argv: dict = None) -> argparse.Namespace:
|
|
if argv is None:
|
|
argv = sys.argv
|
|
|
|
parser = argparse.ArgumentParser(description='Exercise 1')
|
|
parser.add_argument('input', nargs='?', type=argparse.FileType('r'), help='Documents to read.', default=sys.stdin)
|
|
parser.add_argument('similarity', nargs='?', type=float, help='Similarity threshold.', default=0.05)
|
|
|
|
return parser.parse_args(argv[1:])
|
|
|
|
|
|
def normalize(doc: str) -> str:
|
|
"""
|
|
Remove accents from letters, remove non-ascii letters, keep only letters and digits.
|
|
"""
|
|
return ''.join(char for char in unicodedata.normalize(
|
|
'NFKD', doc.casefold().replace('æ', 'ae').replace('œ', 'oe'))
|
|
if unicodedata.category(char) in ['Lu', 'Ll', 'Nd']
|
|
).casefold().encode('ascii', 'ignore').decode('ascii')
|
|
|
|
|
|
def compute_shingles(docs: list[str], single_size: int) -> np.ndarray:
|
|
shingle_matrix = np.zeros((2, len(docs)), dtype=bool)
|
|
shingle_id = {}
|
|
|
|
for doc_id, doc in enumerate(docs):
|
|
char_shing = [doc[i:i + single_size] for i in range(len(doc) - single_size + 1)]
|
|
for sh in char_shing:
|
|
if sh not in shingle_id:
|
|
shingle_id[sh] = len(shingle_id)
|
|
if shingle_id[sh] >= len(shingle_matrix):
|
|
# Extend matrix, double its size
|
|
shingle_matrix = np.append(shingle_matrix, np.zeros(shingle_matrix.shape, dtype=bool), axis=0)
|
|
|
|
shingle_matrix[shingle_id[sh], doc_id] = True
|
|
|
|
shingle_matrix = shingle_matrix[:len(shingle_id)]
|
|
|
|
return shingle_matrix
|
|
|
|
|
|
def min_hash(doc: str, perm: list[str]) -> str:
|
|
for d in perm:
|
|
if d in doc:
|
|
return d
|
|
|
|
|
|
def compute_signature_matrix(shingles: np.ndarray, permutations_count: int) -> np.ndarray:
|
|
shingles_count, docs_count = shingles.shape
|
|
|
|
signature_matrix = np.inf * np.ones((permutations_count, docs_count))
|
|
|
|
for permutation_id in range(permutations_count):
|
|
permutation = np.random.permutation(shingles_count)
|
|
|
|
for doc_id, doc_shingles in enumerate(shingles.T):
|
|
for x in permutation:
|
|
if doc_shingles[x]:
|
|
signature_matrix[permutation_id, doc_id] = x
|
|
break
|
|
|
|
return signature_matrix
|
|
|
|
|
|
def jaccard_similarity(doc1: set, doc2: set) -> float:
|
|
if not doc1 or not doc2:
|
|
return 0.0
|
|
|
|
inter = doc1.intersection(doc2)
|
|
union = doc1.union(doc2)
|
|
return len(inter) / len(union)
|
|
|
|
|
|
def parse(stream, similarity: float) -> None:
|
|
docs = [line.rstrip('\n') for line in stream]
|
|
docs = [normalize(doc) for doc in docs] # Remove special characters and normalize accents
|
|
|
|
shingles = compute_shingles(docs, SHINGLE_SIZE)
|
|
signature = compute_signature_matrix(shingles, PERMUTATIONS_COUNT)
|
|
|
|
shingles_sets = [set(x for x in range(len(shingles)) if shingles[x, doc_id]) for doc_id in range(len(docs))]
|
|
for i, doc1 in enumerate(shingles_sets):
|
|
t = time.time()
|
|
for j in range(i + 1, len(docs)):
|
|
doc2 = shingles_sets[j]
|
|
d = jaccard_similarity(doc1, doc2)
|
|
if d >= similarity:
|
|
print(f"{i} {j} {d:.06f}")
|
|
|
|
|
|
def main():
|
|
ns = parse_args()
|
|
parse(ns.input, ns.similarity)
|