feat: support prefix matching and must_not clause (#2441)

This commit is contained in:
BubbleCal
2025-06-19 10:32:32 +08:00
committed by GitHub
parent c72f6770fd
commit cbb5a841b1
6 changed files with 124 additions and 33 deletions

View File

@@ -101,8 +101,9 @@ class FullTextOperator(str, Enum):
class Occur(str, Enum):
MUST = "MUST"
SHOULD = "SHOULD"
MUST = "MUST"
MUST_NOT = "MUST_NOT"
@pydantic.dataclasses.dataclass
@@ -181,6 +182,9 @@ class MatchQuery(FullTextQuery):
Can be either `AND` or `OR`.
If `AND`, all terms in the query must match.
If `OR`, at least one term in the query must match.
prefix_length : int, optional
The number of beginning characters being unchanged for fuzzy matching.
This is useful to achieve prefix matching.
"""
query: str
@@ -189,6 +193,7 @@ class MatchQuery(FullTextQuery):
fuzziness: int = pydantic.Field(0, kw_only=True)
max_expansions: int = pydantic.Field(50, kw_only=True)
operator: FullTextOperator = pydantic.Field(FullTextOperator.OR, kw_only=True)
prefix_length: int = pydantic.Field(0, kw_only=True)
def query_type(self) -> FullTextQueryType:
return FullTextQueryType.MATCH

View File

@@ -6,7 +6,7 @@ import lancedb
# --8<-- [end:import-lancedb]
# --8<-- [start:import-numpy]
from lancedb.query import BoostQuery, MatchQuery
from lancedb.query import BooleanQuery, BoostQuery, MatchQuery, Occur
import numpy as np
import pyarrow as pa
@@ -191,6 +191,15 @@ def test_fts_fuzzy_query():
"food", # 1 insertion
}
results = table.search(
MatchQuery("foo", "text", fuzziness=1, prefix_length=3)
).to_pandas()
assert len(results) == 2
assert set(results["text"].to_list()) == {
"foo",
"food",
}
@pytest.mark.skipif(
os.name == "nt", reason="Need to fix https://github.com/lancedb/lance/issues/3905"
@@ -240,6 +249,60 @@ def test_fts_boost_query():
)
@pytest.mark.skipif(
os.name == "nt", reason="Need to fix https://github.com/lancedb/lance/issues/3905"
)
def test_fts_boolean_query(tmp_path):
uri = tmp_path / "boolean-example"
db = lancedb.connect(uri)
table = db.create_table(
"my_table_fts_boolean",
data=[
{"text": "The cat and dog are playing"},
{"text": "The cat is sleeping"},
{"text": "The dog is barking"},
{"text": "The dog chases the cat"},
],
mode="overwrite",
)
table.create_fts_index("text", use_tantivy=False, replace=True)
# SHOULD
results = table.search(
MatchQuery("cat", "text") | MatchQuery("dog", "text")
).to_pandas()
assert len(results) == 4
assert set(results["text"].to_list()) == {
"The cat and dog are playing",
"The cat is sleeping",
"The dog is barking",
"The dog chases the cat",
}
# MUST
results = table.search(
MatchQuery("cat", "text") & MatchQuery("dog", "text")
).to_pandas()
assert len(results) == 2
assert set(results["text"].to_list()) == {
"The cat and dog are playing",
"The dog chases the cat",
}
# MUST NOT
results = table.search(
BooleanQuery(
[
(Occur.MUST, MatchQuery("cat", "text")),
(Occur.MUST_NOT, MatchQuery("dog", "text")),
]
)
).to_pandas()
assert len(results) == 1
assert set(results["text"].to_list()) == {
"The cat is sleeping",
}
@pytest.mark.skipif(
os.name == "nt", reason="Need to fix https://github.com/lancedb/lance/issues/3905"
)