forked from pgvector/pgvector-python
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtest_peewee.py
More file actions
223 lines (180 loc) · 9.21 KB
/
test_peewee.py
File metadata and controls
223 lines (180 loc) · 9.21 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
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
214
215
216
217
218
219
220
221
222
223
from math import sqrt
import numpy as np
from peewee import Model, PostgresqlDatabase, fn
from pgvector.peewee import VectorField, HalfVectorField, FixedBitField, SparseVectorField, SparseVector
db = PostgresqlDatabase('pgvector_python_test')
class BaseModel(Model):
class Meta:
database = db
class Item(BaseModel):
embedding = VectorField(dimensions=3, null=True)
half_embedding = HalfVectorField(dimensions=3, null=True)
binary_embedding = FixedBitField(max_length=3, null=True)
sparse_embedding = SparseVectorField(dimensions=3, null=True)
class Meta:
table_name = 'peewee_item'
Item.add_index('embedding vector_l2_ops', using='hnsw')
db.connect()
db.execute_sql('CREATE EXTENSION IF NOT EXISTS vector')
db.drop_tables([Item])
db.create_tables([Item])
def create_items():
Item.create(id=1, embedding=[1, 1, 1], half_embedding=[1, 1, 1], binary_embedding='000', sparse_embedding=SparseVector([1, 1, 1]))
Item.create(id=2, embedding=[2, 2, 2], half_embedding=[2, 2, 2], binary_embedding='101', sparse_embedding=SparseVector([2, 2, 2]))
Item.create(id=3, embedding=[1, 1, 2], half_embedding=[1, 1, 2], binary_embedding='111', sparse_embedding=SparseVector([1, 1, 2]))
class TestPeewee:
def setup_method(self, test_method):
Item.truncate_table()
def test_vector(self):
Item.create(id=1, embedding=[1, 2, 3])
item = Item.get_by_id(1)
assert np.array_equal(item.embedding, np.array([1, 2, 3]))
assert item.embedding.dtype == np.float32
def test_vector_l2_distance(self):
create_items()
distance = Item.embedding.l2_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 3, 2]
assert [v.distance for v in items] == [0, 1, sqrt(3)]
def test_vector_max_inner_product(self):
create_items()
distance = Item.embedding.max_inner_product([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [2, 3, 1]
assert [v.distance for v in items] == [-6, -4, -3]
def test_vector_cosine_distance(self):
create_items()
distance = Item.embedding.cosine_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 2, 3]
assert [v.distance for v in items] == [0, 0, 0.05719095841793653]
def test_vector_l1_distance(self):
create_items()
distance = Item.embedding.l1_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 3, 2]
assert [v.distance for v in items] == [0, 1, 3]
def test_halfvec(self):
Item.create(id=1, half_embedding=[1, 2, 3])
item = Item.get_by_id(1)
assert item.half_embedding.to_list() == [1, 2, 3]
def test_halfvec_l2_distance(self):
create_items()
distance = Item.half_embedding.l2_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 3, 2]
assert [v.distance for v in items] == [0, 1, sqrt(3)]
def test_halfvec_max_inner_product(self):
create_items()
distance = Item.half_embedding.max_inner_product([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [2, 3, 1]
assert [v.distance for v in items] == [-6, -4, -3]
def test_halfvec_cosine_distance(self):
create_items()
distance = Item.half_embedding.cosine_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 2, 3]
assert [v.distance for v in items] == [0, 0, 0.05719095841793653]
def test_halfvec_l1_distance(self):
create_items()
distance = Item.half_embedding.l1_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 3, 2]
assert [v.distance for v in items] == [0, 1, 3]
def test_bit(self):
Item.create(id=1, binary_embedding='101')
item = Item.get_by_id(1)
assert item.binary_embedding == '101'
def test_bit_hamming_distance(self):
create_items()
distance = Item.binary_embedding.hamming_distance('101')
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [2, 3, 1]
assert [v.distance for v in items] == [0, 1, 2]
def test_bit_jaccard_distance(self):
create_items()
distance = Item.binary_embedding.jaccard_distance('101')
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [2, 3, 1]
# assert [v.distance for v in items] == [0, 1/3, 1]
def test_sparsevec(self):
Item.create(id=1, sparse_embedding=[1, 2, 3])
item = Item.get_by_id(1)
assert item.sparse_embedding.to_list() == [1, 2, 3]
def test_sparsevec_l2_distance(self):
create_items()
distance = Item.sparse_embedding.l2_distance(SparseVector([1, 1, 1]))
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 3, 2]
assert [v.distance for v in items] == [0, 1, sqrt(3)]
def test_sparsevec_max_inner_product(self):
create_items()
distance = Item.sparse_embedding.max_inner_product([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [2, 3, 1]
assert [v.distance for v in items] == [-6, -4, -3]
def test_sparsevec_cosine_distance(self):
create_items()
distance = Item.sparse_embedding.cosine_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 2, 3]
assert [v.distance for v in items] == [0, 0, 0.05719095841793653]
def test_sparsevec_l1_distance(self):
create_items()
distance = Item.sparse_embedding.l1_distance([1, 1, 1])
items = Item.select(Item.id, distance.alias('distance')).order_by(distance).limit(5)
assert [v.id for v in items] == [1, 3, 2]
assert [v.distance for v in items] == [0, 1, 3]
def test_where(self):
create_items()
items = Item.select().where(Item.embedding.l2_distance([1, 1, 1]) < 1)
assert [v.id for v in items] == [1]
def test_vector_avg(self):
avg = Item.select(fn.avg(Item.embedding).coerce(True)).scalar()
assert avg is None
Item.create(embedding=[1, 2, 3])
Item.create(embedding=[4, 5, 6])
avg = Item.select(fn.avg(Item.embedding).coerce(True)).scalar()
assert np.array_equal(avg, np.array([2.5, 3.5, 4.5]))
def test_vector_sum(self):
sum = Item.select(fn.sum(Item.embedding).coerce(True)).scalar()
assert sum is None
Item.create(embedding=[1, 2, 3])
Item.create(embedding=[4, 5, 6])
sum = Item.select(fn.sum(Item.embedding).coerce(True)).scalar()
assert np.array_equal(sum, np.array([5, 7, 9]))
def test_halfvec_avg(self):
avg = Item.select(fn.avg(Item.half_embedding).coerce(True)).scalar()
assert avg is None
Item.create(half_embedding=[1, 2, 3])
Item.create(half_embedding=[4, 5, 6])
avg = Item.select(fn.avg(Item.half_embedding).coerce(True)).scalar()
assert avg.to_list() == [2.5, 3.5, 4.5]
def test_halfvec_sum(self):
sum = Item.select(fn.sum(Item.half_embedding).coerce(True)).scalar()
assert sum is None
Item.create(half_embedding=[1, 2, 3])
Item.create(half_embedding=[4, 5, 6])
sum = Item.select(fn.sum(Item.half_embedding).coerce(True)).scalar()
assert sum.to_list() == [5, 7, 9]
def test_get_or_create(self):
Item.get_or_create(id=1, defaults={'embedding': [1, 2, 3]})
Item.get_or_create(embedding=np.array([4, 5, 6]))
Item.get_or_create(embedding=Item.embedding.to_value([7, 8, 9]))
def test_vector_array(self):
from playhouse.postgres_ext import PostgresqlExtDatabase, ArrayField
ext_db = PostgresqlExtDatabase('pgvector_python_test')
class ExtItem(BaseModel):
embeddings = ArrayField(VectorField, field_kwargs={'dimensions': 3}, index=False)
class Meta:
database = ext_db
table_name = 'peewee_ext_item'
ext_db.connect()
ext_db.drop_tables([ExtItem])
ext_db.create_tables([ExtItem])
# fails with column "embeddings" is of type vector[] but expression is of type text[]
# ExtItem.create(id=1, embeddings=[np.array([1, 2, 3]), np.array([4, 5, 6])])
# item = ExtItem.get_by_id(1)
# assert np.array_equal(item.embeddings[0], np.array([1, 2, 3]))
# assert np.array_equal(item.embeddings[1], np.array([4, 5, 6]))