def find_near_duplicate(dataset, query, targets, labels, min_jaccard_value, no_of_bands, n_permutations, n_gram, n_gram_type='char'):
"""Using LSH object finds the near duplicate strings.
Args:
query_sentences (dict): Dict with query strings and version of string in lower case and without comma.
sentences (dict): Dict with target strings and version of string in lower case and without comma.
min_jaccard_value (float): Minimum value for the Jaccard Distance.
no_of_bands (int): Number of bands to break minhash signature into before hashing into buckets.
n_permutations (int): Number of permutations used to create minhash signatures used in LSH model.
n_gram (int): Size of each overlapping text shingle to break text into prior to hashing.
"""
for i in targets:
if len(i) < n_gram:
pp(i)
# Create MinHash object.
minhash = MinHash(targets, n_gram=n_gram, n_gram_type=n_gram_type,
permutations=n_permutations, hash_bits=64, seed=SEED)
# Create LSH model.
lsh = LSH(minhash, labels, no_of_bands=no_of_bands)
# Query to find near duplicates the string in `search`
closest_results = lsh.query(labels[0], min_jaccard=min_jaccard_value)
# print("QUERY: {}".format(labels[0]))
# pp(closest_results)
return {"dataset": dataset, "query": labels[0], "duplicates": ' '.join(closest_results)}
def test_lsh_errors():
with pytest.raises(ValueError):
LSH(content)
with pytest.raises(ValueError):
LSH(labels=labels)
with pytest.raises(ValueError):
LSH(minhash, labels, no_of_bands=49)
def find_adjacency(draws):
draws_idx = list(draws)
draws_nos = list(draws.values())
# labels = [draws_idx[0]]
# content = [" ".join(map(str, draws_nos[0]))]
# print(labels)
# print(content)
# minhash = MinHash(content, n_gram=9, permutations=100, hash_bits=64, seed=3)
# lsh = LSH(minhash, labels, no_of_bands=50)
# print(lsh.query(1, min_jaccard=0.5))
new_labels = []
new_content = []
for i in range(0, len(draws)):
new_labels.append(draws_idx[i])
new_content.append(" ".join(map(str, draws_nos[i])))
# print(new_labels)
# print(new_content)
new_labels.append(99999)
new_content.append(" ".join(map(str, draws_nos[0])))
minhash = MinHash(new_content,
n_gram=9,
permutations=500,
hash_bits=64,
seed=3)
lsh = LSH(minhash, new_labels, no_of_bands=100)
adjacency_list = lsh.adjacency_list(min_jaccard=0.5)
for key, value in adjacency_list.items():
if len(value) > 0:
print(key, value)
class LshCluster():
def __init__(self, Content, nGram=3, PrenutNum=16, BandNum=8, MinJaccard=0.2):
self.nGram = nGram
self.BandNum = BandNum
self.PrenutNum = PrenutNum
self.MinJaccard = MinJaccard
self.Seed = 3
self.CreateLsh (Content)
def Transform(self, Contexts):
NewContexts = []
for ctx in Contexts:
ctx = ctx.strip()
ctx = ctx.replace("_", "")
ctx = ctx.lower()
NewContexts.append(ctx)
return NewContexts
def CreateLsh(self, InContext):
Labels = range(len(InContext))
InContext = self.Transform (InContext)
Hash = MinHash(InContext, n_gram=self.nGram, permutations=self.PrenutNum, hash_bits=64, seed=self.Seed)
self.Lsh = LSH(Hash, Labels, no_of_bands=self.BandNum)
self.MaxIndex = len (self.Lsh.adjacency_list(min_jaccard=self.MinJaccard))
def QuerySimilars(self, Index):
if Index >= self.MaxIndex:
return []
Results = self.Lsh.query(Index, min_jaccard=self.MinJaccard)
return Results
def test_initialize_from_empty_lsh():
lsh = LSH()
assert lsh.no_of_bands is None
assert lsh._buckets == defaultdict(list)
assert lsh._i_bucket == defaultdict(list)
assert lsh.permutations is None
lsh.update(minhash, labels)
assert list(lsh._i_bucket) == labels
assert lsh.permutations == 100
assert lsh.no_of_bands == 50
def test_initialize_from_empty_lsh():
lsh = LSH()
assert lsh.no_of_bands is None
assert lsh._buckets == defaultdict(list)
assert lsh._i_bucket == defaultdict(list)
assert lsh.permutations is None
lsh.update(minhash, labels)
assert list(lsh._i_bucket) == labels
buckets = lsh._buckets
assert buckets[4466445138223010106] == [1, 8]
assert buckets[-3939654010681976230] == [1, 4, 8]
assert lsh.permutations == 100
assert lsh.no_of_bands == 50
def create_lsh(self, content, no_of_bands, n_permutations, n_gram):
"""Create Minhash and Locality Sensitive Hashing (LSH) to detect near duplicate texts.
Args:
content (list): List with string to build LSH.
no_of_bands (int): Number of bands to break minhash signature into before hashing into buckets.
n_permutations (int): Number of permutations used to create minhash signatures used in LSH model.
n_gram (int): Size of each overlapping text shingle to break text into prior to hashing.
no_of_bands(int): Number of bands to break minhash signature into before hashing into buckets.
Returns:
class 'snapy.lsh.LSH': Snapy LSH object.
"""
labels = range(len(content))
# Create MinHash object.
minhash = MinHash(content,
n_gram=n_gram,
permutations=n_permutations,
hash_bits=64,
seed=SEED)
# Create LSH model.
lsh = LSH(minhash, labels, no_of_bands=no_of_bands)
return lsh
def test_lsh_edge_list():
lsh = LSH(minhash, labels)
with pytest.raises(ValueError):
lsh.edge_list(sensitivity=101)
assert lsh.edge_list() == [(8, 1), (8, 4), (5, 3), (4, 1)]
assert lsh.edge_list(sensitivity=20) == [(8, 1), (5, 3), (4, 1)]
assert lsh.edge_list(min_jaccard=0.7) == []
assert lsh.edge_list(min_jaccard=0.6) == [(5, 3)]
assert lsh.edge_list(jaccard_weighted=True,
min_jaccard=0.55) == [(5, 3, 0.6), (4, 1, 0.58)]
def get_lsh_model(documents, seed_int):
seed = seed_int
# Create MinHash object.
minhash = MinHash(my_documents,
n_gram=9,
permutations=100,
hash_bits=64,
seed=seed)
# Create LSH model.
lsh = LSH(minhash, labels, no_of_bands=50)
return lsh
def test_update_lsh():
lsh = LSH(minhash, labels)
with pytest.raises(ValueError):
lsh.update(minhash, labels)
new_content = [
'Jupiter is primarily composed of hydrogen with a quarter of its mass being helium',
'Jupiter moving out of the inner Solar System would have allowed the formation of inner planets.'
]
new_labels = [11, 12]
incorrect_minhash = MinHash(new_content, permutations=10)
with pytest.raises(ValueError):
lsh.update(incorrect_minhash, new_labels)
correct_minhash = MinHash(new_content)
lsh.update(correct_minhash, new_labels)
assert lsh.permutations == 100
assert list(lsh._i_bucket) == labels + [11, 12]
def _get_duplicate_ids(text: List[str], lsh: LSH,
min_jaccard: float) -> Iterable[str]:
"""Uses the given `lsh` object to find near duplicate text in `text`. Returns a list of
indices into `text` which point to duplicate texts.
"""
duplicate_ids = set()
adjacency_list = lsh.adjacency_list(min_jaccard=min_jaccard)
with typer.progressbar(adjacency_list.items(),
label="Deduplicating text") as progress:
for query_id, similar_ids in progress:
# If query_id exists in duplicate_ids, we have already accounted for it.
if query_id in duplicate_ids:
continue
duplicate_ids.update(similar_ids)
typer.secho(
f"{SEARCH} Found a total of {len(duplicate_ids)} duplicate texts.",
bold=True,
)
return list(duplicate_ids)
def test_lsh_adjacency_list():
lsh = LSH(minhash, labels)
with pytest.raises(ValueError):
lsh.adjacency_list(sensitivity=1000)
sensitivity_list = lsh.adjacency_list(sensitivity=2)
assert sensitivity_list == {
1: [8, 4],
2: [],
3: [5],
4: [1, 8],
5: [3],
6: [],
7: [],
8: [1, 4],
9: []
}
jaccard_list = lsh.adjacency_list(min_jaccard=0.6)
assert jaccard_list == {
1: [],
2: [],
3: [5],
4: [],
5: [3],
6: [],
7: [],
8: [],
9: []
}
default_list = lsh.adjacency_list()
assert default_list == {
1: [8, 4],
2: [],
3: [5],
4: [1, 8],
5: [3],
6: [],
7: [],
8: [1, 4],
9: []
}
def _create_lsh(
text: List[str],
labels: List[int],
n_gram: int,
n_permutations: int,
hash_bits: int,
no_of_bands: int,
) -> LSH:
"""Returns a `snapy.lsh.LSH` object constructed from `text` to detect near duplicate texts.
"""
minhash = MinHash(text,
n_gram=n_gram,
permutations=n_permutations,
hash_bits=hash_bits,
seed=SEED)
lsh = LSH(minhash, labels, no_of_bands=no_of_bands)
typer.secho(
f"{HASHING} Hashed the normalized text using Locality-Sensitive Hashing (LSH).",
bold=True,
)
return lsh
def identify_dublicates(self, ctnt_to_dedup):
_ix = [i for i in range(len(ctnt_to_dedup))]
_mn_hash = MinHash(ctnt_to_dedup, n_gram=self.n_gram, seed=self.seed)
_lsh = LSH(_mn_hash, _ix, no_of_bands=self.lsh_bands)
candidates = _lsh.adjacency_list(min_jaccard=self.j_thresh)
return candidates
def test_lsh_remove():
lsh = LSH(minhash, labels)
lsh.remove(5)
assert list(lsh._i_bucket) == [1, 2, 3, 4, 6, 7, 8, 9]
with pytest.raises(KeyError):
lsh.remove(11)
def test_lsh_contains():
lsh = LSH(minhash, labels)
assert lsh.contains() == labels
def test_lsh_query():
lsh = LSH(minhash, labels)
with pytest.raises(KeyError):
lsh.query(10)
with pytest.raises(KeyError):
lsh.query(0)
with pytest.raises(ValueError):
lsh.query(2, sensitivity=100)
result = lsh.query(1)
assert result == [8, 4]
result = lsh.query(1, sensitivity=29)
assert result == [4]
result = lsh.query(1, min_jaccard=0.55)
assert result == [4]
def CreateLsh(self, InContext):
Labels = range(len(InContext))
InContext = self.Transform (InContext)
Hash = MinHash(InContext, n_gram=self.nGram, permutations=self.PrenutNum, hash_bits=64, seed=self.Seed)
self.Lsh = LSH(Hash, Labels, no_of_bands=self.BandNum)
self.MaxIndex = len (self.Lsh.adjacency_list(min_jaccard=self.MinJaccard))
from snapy import MinHash, LSH
import numpy as np
from fasta_parser import parse_to_list
contigs_file = "../contigs-outputs/basic_k-mer24/basic_try_k-mer24.contigs.fa"
print("parsing contigs to list...")
contigs_list, num_contigs = parse_to_list(contigs_file)
print("number of contigs (shorter then 1,00bp):", num_contigs)
labels = np.arange(num_contigs).tolist()
# Create MinHash object.
print("creating minhash object...")
minhash = MinHash(contigs_list, n_gram=24)
# Create LSH model.
print("creating LSH object...")
lsh = LSH(minhash, labels)
print("query object:")
print("similar strings to the string in index 0:", lsh.query(0))
print("similar strings to the string in index 1:", lsh.query(1))
print("similar strings to the string in index 2:", lsh.query(2))
print("similar strings to the string in index 3:", lsh.query(3))
print("similar strings to the string in index 4:", lsh.query(4))
print("similar strings to the string in index 5:", lsh.query(5))
print(lsh.contains())
#number of randomly sampled hash values to use for generating each texts minhash signature (larger = more accurate & slower)
permutations = 100
#hash value size to be used to generate minhash signitures from shingles (32,64, or 128 bit).
#NOTE: should be chosen based on text length and a trade off between performance ad accuracy
hash_bits = 64
# Create MinHash object.
minhash = MinHash(content,
n_gram=n_gram,
permutations=permutations,
hash_bits=hash_bits,
seed=seed)
# Create LSH model.
lsh = LSH(minhash, labels, no_of_bands=50)
#query to find near duplicates for text 1
print(lsh.query(1, min_jaccard=.5))
#update model
#generate minhash aignitures for new text, and add new texts to LSH model
new_text = [
'Jupiter is primarily composed of hydrogen with a quarter of its mass being helium',
'Jupiter moving out of the inner Solar System would have allowed the formation of inner planets.',
]
new_labels = ['new_doc1', 'new_doc2']
#1.create minhash signitues for new text
new_minhash = MinHash(new_text,
def test_initialize_lsh_with_params():
lsh = LSH(minhash, labels, no_of_bands=20)
assert lsh.no_of_bands == 20
assert lsh.permutations == 100
assert list(lsh._i_bucket) == labels
