99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213 | class MinHashLSHBlocker(SchemaAgnosticBlocker):
"""Blocker relying on MinHashLSH procedure.
Args:
----
tokenize_fn Callable: Function that tokenizes entity attribute values.
threshold: float: Jaccard threshold to use in underlying lsh procedure.
num_perm: int: number of permutations used in minhash algorithm.
weights: Tuple[float,float]: false positive/false negative weighting (must add up to one)
Attributes:
----------
tokenize_fn Callable: Function that tokenizes entity attribute values.
threshold: float: Jaccard threshold to use in underlying lsh procedure.
num_perm: int: number of permutations used in minhash algorithm.
weights: Tuple[float,float]: false positive/false negative weighting (must add up to one)
Examples:
--------
>>> # doctest: +SKIP
>>> from sylloge import MovieGraphBenchmark
>>> from klinker.data import KlinkerDataset
>>> ds = KlinkerDataset.from_sylloge(MovieGraphBenchmark(),clean=True)
>>> from klinker.blockers import MinHashLSHBlocker
>>> blocker = MinHashLSHBlocker(threshold=0.8, weights=(0.7,0.3))
>>> blocks = blocker.assign(left=ds.left, right=ds.right)
"""
def __init__(
self,
tokenize_fn: Callable = word_tokenize,
stop_words: Optional[List[str]] = None,
min_token_length: int = 3,
threshold: float = 0.5,
num_perm: int = 128,
weights: Tuple[float, float] = (0.5, 0.5),
):
self.tokenizer = FilteredTokenizer(
tokenize_fn=tokenize_fn,
min_token_length=min_token_length,
stop_words=stop_words,
)
self.threshold = threshold
self.num_perm = num_perm
self.weights = weights
def _inner_encode(self, val: str):
"""Encodes string to list of bytes.
Args:
----
val: str: input string.
Returns:
-------
list of bytes.
"""
return [tok.encode("utf-8") for tok in self.tokenizer.tokenize(str(val))]
def _assign(
self,
left: SeriesType,
right: SeriesType,
left_rel: Optional[KlinkerFrame] = None,
right_rel: Optional[KlinkerFrame] = None,
) -> KlinkerBlockManager:
"""Assign entity ids to blocks.
Uses minhash algorithm to encode entities via tokenized attributes.
Fills a lsh instance with the left hashes.
Queries using the right hashes.
Args:
----
left: SeriesType: concatenated entity attribute values of left dataset as series.
right: SeriesType: concatenated entity attribute values of left dataset as series.
left_rel: Optional[KlinkerFrame]: (Default value = None) Contains relational information of left dataset.
right_rel: Optional[KlinkerFrame]: (Default value = None) Contains relational information of left dataset.
Returns:
-------
KlinkerBlockManager: instance holding the resulting blocks.
"""
lsh = MinHashLSH(
threshold=self.threshold,
num_perm=self.num_perm,
weights=self.weights,
)
if isinstance(left, dd.Series):
left.map_partitions(
_insert,
lsh=lsh,
encode_fn=self._inner_encode,
meta=left._meta.index,
).compute()
blocks = right.map_partitions(
_query,
lsh=lsh,
encode_fn=self._inner_encode,
left_name=left.name,
right_name=right.name,
meta=pd.DataFrame([], columns=[left.name, right.name], dtype="O"),
)
return KlinkerBlockManager(blocks)
else:
_insert(left, lsh=lsh, encode_fn=self._inner_encode)
blocks = _query(
right,
lsh=lsh,
encode_fn=self._inner_encode,
left_name=left.name,
right_name=right.name,
)
return KlinkerBlockManager.from_pandas(blocks)
|