-
Notifications
You must be signed in to change notification settings - Fork 237
Expand file tree
/
Copy pathfind.py
More file actions
185 lines (160 loc) · 5.62 KB
/
find.py
File metadata and controls
185 lines (160 loc) · 5.62 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
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
from typing import (
TYPE_CHECKING,
TypeVar,
Sequence,
List,
Union,
Optional,
Dict,
)
import numpy as np
from docarray import Document, DocumentArray
from docarray.math import ndarray
from docarray.math.helper import EPSILON
from docarray.math.ndarray import to_numpy_array
from docarray.score import NamedScore
from docarray.array.mixins.find import FindMixin as BaseFindMixin
if TYPE_CHECKING: # pragma: no cover
import tensorflow
import torch
ElasticArrayType = TypeVar(
'ElasticArrayType',
np.ndarray,
tensorflow.Tensor,
torch.Tensor,
Sequence[float],
Dict,
)
class FindMixin(BaseFindMixin):
def _find_similar_vectors(
self,
query: 'ElasticArrayType',
filter: Optional[Dict] = None,
limit=10,
**kwargs,
):
"""
Return vector search results for the input query. `script_score` will be used in filter_field is set.
:param query: query vector used for vector search
:param filter: filter query used for pre-filtering
:param limit: number of items to be retrieved
:return: DocumentArray containing the closest documents to the query if it is a single query, otherwise a list of DocumentArrays containing
the closest Document objects for each of the queries in `query`.
"""
query = to_numpy_array(query)
is_all_zero = np.all(query == 0)
if is_all_zero:
query = query + EPSILON
knn_query = {
'field': 'embedding',
'query_vector': query,
'k': limit,
'num_candidates': 10000
if 'num_candidates' not in kwargs
else kwargs['num_candidates'],
}
resp = self._client.knn_search(
index=self._config.index_name,
knn=knn_query,
filter=filter,
)
list_of_hits = resp['hits']['hits']
da = DocumentArray()
for result in list_of_hits:
doc = Document.from_base64(result['_source']['blob'])
doc.scores['score'] = NamedScore(value=result['_score'])
doc.embedding = result['_source']['embedding']
da.append(doc)
return da
def _find_similar_documents_from_text(
self,
query: str,
index: str = 'text',
filter: Union[dict, list] = None,
limit: int = 10,
):
"""
Return keyword matches for the input query
:param query: text used for keyword search
:param limit: number of items to be retrieved
:return: DocumentArray containing the closest documents to the query if it is a single query, otherwise a list of DocumentArrays containing
the closest Document objects for each of the queries in `query`.
"""
query = {
"bool": {
"must": [
{"match": {index: query}},
],
"filter": filter,
}
}
resp = self._client.search(
index=self._config.index_name,
query=query,
source=['id', 'blob', 'text'],
size=limit,
)
list_of_hits = resp['hits']['hits']
da = DocumentArray()
for result in list_of_hits[:limit]:
doc = Document.from_base64(result['_source']['blob'])
doc.scores['score'] = NamedScore(value=result['_score'])
da.append(doc)
return da
def _find_by_text(
self,
query: Union[str, List[str]],
index: str = 'text',
filter: Union[dict, list] = None,
limit: int = 10,
):
if isinstance(query, str):
query = [query]
return [
self._find_similar_documents_from_text(
q,
index=index,
filter=filter,
limit=limit,
)
for q in query
]
def _find(
self,
query: 'ElasticArrayType',
limit: int = 10,
filter: Optional[Dict] = None,
**kwargs,
) -> List['DocumentArray']:
"""Returns approximate nearest neighbors given a batch of input queries.
:param query: input supported to be stored in Elastic. This includes any from the list '[np.ndarray, tensorflow.Tensor, torch.Tensor, Sequence[float]]'
:param limit: number of retrieved items
:param filter: filter query used for pre-filtering
:return: DocumentArray containing the closest documents to the query if it is a single query, otherwise a list of DocumentArrays containing
the closest Document objects for each of the queries in `query`.
"""
query = np.array(query)
num_rows, n_dim = ndarray.get_array_rows(query)
if n_dim != 2:
query = query.reshape((num_rows, -1))
return [
self._find_similar_vectors(q, filter=filter, limit=limit, **kwargs)
for q in query
]
def _find_with_filter(self, query: Dict, limit: Optional[Union[int, float]] = 20):
resp = self._client.search(
index=self._config.index_name,
query=query,
size=limit,
)
list_of_hits = resp['hits']['hits']
da = DocumentArray()
for result in list_of_hits[:limit]:
doc = Document.from_base64(result['_source']['blob'])
doc.scores['score'] = NamedScore(value=result['_score'])
da.append(doc)
return da
def _filter(
self, query: Dict, limit: Optional[Union[int, float]] = 20
) -> 'DocumentArray':
return self._find_with_filter(query, limit=limit)