def find_near_matches_levenshtein(subsequence, sequence, max_l_dist):
"""Find near-matches of the subsequence in the sequence.
This chooses a suitable fuzzy search implementation according to the given
parameters.
Returns a list of fuzzysearch.Match objects describing the matching parts
of the sequence.
"""
if not subsequence:
raise ValueError('Given subsequence is empty!')
if max_l_dist < 0:
raise ValueError('Maximum Levenshtein distance must be >= 0!')
if max_l_dist == 0:
return [
Match(start_index, start_index + len(subsequence), 0)
for start_index in search_exact(subsequence, sequence)
]
elif len(subsequence) // (max_l_dist + 1) >= 3:
return find_near_matches_levenshtein_ngrams(subsequence, sequence,
max_l_dist)
else:
matches = find_near_matches_levenshtein_linear_programming(
subsequence, sequence, max_l_dist)
match_groups = group_matches(matches)
best_matches = [
get_best_match_in_group(group) for group in match_groups
]
return sorted(best_matches)
def _find_near_matches_generic_ngrams(subsequence, sequence, search_params):
max_substitutions, max_insertions, max_deletions, max_l_dist = search_params.unpacked
# optimization: prepare some often used things in advance
subseq_len = len(subsequence)
seq_len = len(sequence)
ngram_len = subseq_len // (max_l_dist + 1)
if ngram_len == 0:
raise ValueError(
'the subsequence length must be greater than max_l_dist')
for ngram_start in xrange(0, subseq_len - ngram_len + 1, ngram_len):
ngram_end = ngram_start + ngram_len
start_index = max(0, ngram_start - max_l_dist)
end_index = min(seq_len, seq_len - subseq_len + ngram_end + max_l_dist)
for index in search_exact(subsequence[ngram_start:ngram_end], sequence,
start_index, end_index):
# try to expand left and/or right according to n_ngram
for match in find_near_matches_generic_linear_programming(
subsequence,
sequence[max(0, index - ngram_start - max_l_dist):index -
ngram_start + subseq_len + max_l_dist],
search_params,
):
yield match._replace(
start=match.start +
max(0, index - ngram_start - max_l_dist),
end=match.end + max(0, index - ngram_start - max_l_dist),
)
def find_near_matches_substitutions(subsequence, sequence, max_substitutions):
"""Find near-matches of the subsequence in the sequence.
This chooses a suitable fuzzy search implementation according to the given
parameters.
Returns a list of fuzzysearch.Match objects describing the matching parts
of the sequence.
"""
_check_arguments(subsequence, sequence, max_substitutions)
if max_substitutions == 0:
return [
Match(start_index, start_index + len(subsequence), 0)
for start_index in search_exact(subsequence, sequence)
]
elif len(subsequence) // (max_substitutions + 1) >= 3:
return find_near_matches_substitutions_ngrams(
subsequence,
sequence,
max_substitutions,
)
else:
return find_near_matches_substitutions_lp(
subsequence,
sequence,
max_substitutions,
)
def find_near_matches_levenshtein_ngrams(subsequence, sequence, max_l_dist):
subseq_len = len(subsequence)
seq_len = len(sequence)
ngram_len = subseq_len // (max_l_dist + 1)
if ngram_len == 0:
raise ValueError(
'the subsequence length must be greater than max_l_dist')
matches = []
for ngram_start in xrange(0, subseq_len - ngram_len + 1, ngram_len):
ngram_end = ngram_start + ngram_len
subseq_before_reversed = subsequence[:ngram_start][::-1]
subseq_after = subsequence[ngram_end:]
start_index = max(0, ngram_start - max_l_dist)
end_index = min(seq_len, seq_len - subseq_len + ngram_end + max_l_dist)
for index in search_exact(subsequence[ngram_start:ngram_end], sequence,
start_index, end_index):
# try to expand left and/or right according to n_ngram
dist_right, right_expand_size = _expand(
subseq_after,
sequence[index + ngram_len:index - ngram_start + subseq_len +
max_l_dist],
max_l_dist,
)
if dist_right is None:
continue
dist_left, left_expand_size = _expand(
subseq_before_reversed,
sequence[max(0, index - ngram_start -
(max_l_dist - dist_right)):index][::-1],
max_l_dist - dist_right,
)
if dist_left is None:
continue
assert dist_left + dist_right <= max_l_dist
matches.append(
Match(
start=index - left_expand_size,
end=index + ngram_len + right_expand_size,
dist=dist_left + dist_right,
))
# don't return overlapping matches; instead, group overlapping matches
# together and return the best match from each group
match_groups = group_matches(matches)
best_matches = [get_best_match_in_group(group) for group in match_groups]
return sorted(best_matches)
def _find_near_matches_substitutions_ngrams(subsequence, sequence,
max_substitutions):
subseq_len = len(subsequence)
seq_len = len(sequence)
ngram_len = subseq_len // (max_substitutions + 1)
if ngram_len == 0:
raise ValueError(
"The subsequence's length must be greater than max_substitutions!")
for ngram_start in range(0, len(subsequence) - ngram_len + 1, ngram_len):
ngram_end = ngram_start + ngram_len
subseq_before = subsequence[:ngram_start]
subseq_after = subsequence[ngram_end:]
for index in search_exact(
subsequence[ngram_start:ngram_end],
sequence,
ngram_start,
seq_len - (subseq_len - ngram_end),
):
n_substitutions = 0
seq_before = sequence[index - ngram_start:index]
if subseq_before != seq_before:
n_substitutions += count_differences_with_maximum(
seq_before, subseq_before,
max_substitutions - n_substitutions + 1)
if n_substitutions > max_substitutions:
continue
seq_after = sequence[index + ngram_len:index - ngram_start +
subseq_len]
if subseq_after != seq_after:
if n_substitutions == max_substitutions:
continue
n_substitutions += count_differences_with_maximum(
seq_after, subseq_after,
max_substitutions - n_substitutions + 1)
if n_substitutions > max_substitutions:
continue
yield Match(
start=index - ngram_start,
end=index - ngram_start + subseq_len,
dist=n_substitutions,
)
def has_near_match_substitutions(subsequence, sequence, max_substitutions):
_check_arguments(subsequence, sequence, max_substitutions)
if max_substitutions == 0:
for start_index in search_exact(subsequence, sequence):
return True
return False
elif len(subsequence) // (max_substitutions + 1) >= 3:
return has_near_match_substitutions_ngrams(
subsequence,
sequence,
max_substitutions,
)
else:
return has_near_match_substitutions_lp(
subsequence,
sequence,
max_substitutions,
)
def choose_search_func(search_params):
max_substitutions, max_insertions, max_deletions, max_l_dist = search_params.unpacked
# if the limitations are so strict that only exact matches are allowed,
# use search_exact()
if search_params.max_l_dist == 0:
return lambda subsequence, sequence, search_params: [
Match(index, index + len(subsequence), 0)
for index in search_exact(subsequence, sequence)
]
# return [
# Match(start_index, start_index + len(subsequence), 0)
# for start_index in search_exact(subsequence, sequence)
# ]
# if only substitutions are allowed, use find_near_matches_substitutions()
elif max_insertions == 0 and max_deletions == 0:
# max_subs = \
# min([x for x in [max_l_dist, max_substitutions] if x is not None])
return lambda subsequence, sequence, search_params:\
find_near_matches_substitutions(
subsequence, sequence,
min([x for x in [search_params.max_l_dist, search_params.max_substitutions] if x is not None])
)
# if it is enough to just take into account the maximum Levenshtein
# distance, use find_near_matches_levenshtein()
elif max_l_dist <= min(
(max_substitutions if max_substitutions is not None else (1 << 29)),
(max_insertions if max_insertions is not None else (1 << 29)),
(max_deletions if max_deletions is not None else (1 << 29)),
):
return lambda subsequence, sequence, search_params:\
find_near_matches_levenshtein(subsequence, sequence, search_params.max_l_dist)
# if none of the special cases above are met, use the most generic version
else:
return find_near_matches_generic
def find_near_matches_generic(subsequence, sequence, search_params):
"""search for near-matches of subsequence in sequence
This searches for near-matches, where the nearly-matching parts of the
sequence must meet the following limitations (relative to the subsequence):
* the maximum allowed number of character substitutions
* the maximum allowed number of new characters inserted
* and the maximum allowed number of character deletions
* the total number of substitutions, insertions and deletions
"""
if not subsequence:
raise ValueError('Given subsequence is empty!')
# if the limitations are so strict that only exact matches are allowed,
# use search_exact()
if search_params.max_l_dist == 0:
return [
Match(start_index, start_index + len(subsequence), 0)
for start_index in search_exact(subsequence, sequence)
]
# if the n-gram length would be at least 3, use the n-gram search method
elif len(subsequence) // (search_params.max_l_dist + 1) >= 3:
return find_near_matches_generic_ngrams(subsequence, sequence,
search_params)
# use the linear programming search method
else:
matches = find_near_matches_generic_linear_programming(
subsequence, sequence, search_params)
match_groups = group_matches(matches)
best_matches = [
get_best_match_in_group(group) for group in match_groups
]
return sorted(best_matches)
def search(self, subsequence, sequence, start_index=0, end_index=None):
return list(search_exact(subsequence, sequence, start_index,
end_index))
def find_near_matches_no_deletions_ngrams(subsequence, sequence,
search_params):
"""search for near-matches of subsequence in sequence
This searches for near-matches, where the nearly-matching parts of the
sequence must meet the following limitations (relative to the subsequence):
* the number of character substitutions must be less than max_substitutions
* no deletions or insertions are allowed
"""
if not subsequence:
raise ValueError('Given subsequence is empty!')
max_substitutions, max_insertions, max_deletions, max_l_dist = search_params.unpacked
max_substitutions = min(max_substitutions, max_l_dist)
max_insertions = min(max_insertions, max_l_dist)
subseq_len = len(subsequence)
seq_len = len(sequence)
ngram_len = subseq_len // (max_substitutions + max_insertions + 1)
if ngram_len == 0:
raise ValueError(
"The subsequence's length must be greater than max_subs + max_ins!"
)
matches = []
matched_indexes = set()
for ngram_start in range(0, len(subsequence) - ngram_len + 1, ngram_len):
ngram_end = ngram_start + ngram_len
subseq_before = subsequence[:ngram_start]
subseq_before_reversed = subseq_before[::-1]
subseq_after = subsequence[ngram_end:]
start_index = max(0, ngram_start - max_insertions)
end_index = min(seq_len,
seq_len - (subseq_len - ngram_end) + max_insertions)
for index in search_exact(
subsequence[ngram_start:ngram_end],
sequence,
start_index,
end_index,
):
if index - ngram_start in matched_indexes:
continue
seq_after = sequence[index + ngram_len:index + subseq_len -
ngram_start + max_insertions]
if seq_after.startswith(subseq_after):
matches_after = [(0, 0)]
else:
matches_after = _expand(subseq_after, seq_after,
max_substitutions, max_insertions,
max_l_dist)
if not matches_after:
continue
_max_substitutions = max_substitutions - min(
m[0] for m in matches_after)
_max_insertions = max_insertions - min(m[1] for m in matches_after)
_max_l_dist = max_l_dist - min(m[0] + m[1] for m in matches_after)
seq_before = sequence[index - ngram_start - _max_insertions:index]
if seq_before.endswith(subseq_before):
matches_before = [(0, 0)]
else:
matches_before = _expand(
subseq_before_reversed,
seq_before[::-1],
_max_substitutions,
_max_insertions,
_max_l_dist,
)
for (subs_before, ins_before) in matches_before:
for (subs_after, ins_after) in matches_after:
if (subs_before + subs_after <= max_substitutions
and ins_before + ins_after <= max_insertions
and subs_before + subs_after + ins_before +
ins_after <= max_l_dist):
matches.append(
Match(
start=index - ngram_start - ins_before,
end=index - ngram_start + subseq_len +
ins_after,
dist=subs_before + subs_after + ins_before +
ins_after,
))
matched_indexes |= set(
range(
index - ngram_start - ins_before,
index - ngram_start - ins_before +
max_insertions + 1,
))
return sorted(matches, key=lambda match: match.start)