pgvector 0.5.0__tar.gz → 0.5.1__tar.gz

{pgvector-0.5.0 → pgvector-0.5.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pgvector
-Version: 0.5.0
+Version: 0.5.1
 Summary: pgvector support for Python
 Author-email: Andrew Kane <andrew@ankane.org>
 License-Expression: MIT

{pgvector-0.5.0 → pgvector-0.5.1}/pgvector/bit.py

@@ -13,7 +13,7 @@ class Bit:
 
     def __init__(
         self,
-        value: bytes | str | list[bool] | np.ndarray[tuple[int, ...], np.dtype[np.bool | np.uint8]],
+        value: bytes | str | list[bool] | np.ndarray[tuple[int, ...], np.dtype[np.bool_ | np.uint8]],
         /
     ) -> None:
         if isinstance(value, bytes):
@@ -26,20 +26,19 @@ class Bit:
                     value = ''.join([bits[v] for v in value])
                 except (KeyError, TypeError):
                     raise ValueError('expected list[bool]')
+            elif not set(value).issubset({'0', '1'}):
+                raise ValueError('expected bit string')
 
             length = len(value)
             if length % 8 != 0:
                 value += '0' * (8 - (length % 8))
 
             self._length = length
-            try:
-                self._data = int(value, 2).to_bytes(len(value) // 8, byteorder='big')
-            except ValueError:
-                raise ValueError('expected bit string')
+            self._data = int(value, 2).to_bytes(len(value) // 8, byteorder='big')
         elif is_ndarray(value):
             import numpy as np
 
-            if value.dtype != np.bool:
+            if value.dtype != np.bool_:
                 # skip error for result of np.unpackbits
                 if value.dtype != np.uint8 or np.any(value > 1):
                     raise ValueError('expected elements to be boolean')
@@ -65,7 +64,7 @@ class Bit:
         # TODO improve
         return [v != '0' for v in self.to_text()]
 
-    def to_numpy(self) -> np.ndarray[tuple[int, ...], np.dtype[np.bool]]:
+    def to_numpy(self) -> np.ndarray[tuple[int, ...], np.dtype[np.bool_]]:
         import numpy as np
 
         return np.unpackbits(np.frombuffer(self._data, dtype=np.uint8), count=self._length).astype(bool)

{pgvector-0.5.0 → pgvector-0.5.1}/pgvector/sparsevec.py

@@ -105,6 +105,11 @@ class SparseVector:
     def _from_sparse(self, arr: sparray | spmatrix, /) -> None:
         value: coo_array | coo_matrix = arr.tocoo(copy=False)  # type: ignore
 
+        # has_canonical_format added in scipy 1.12+
+        if not hasattr(value, 'has_canonical_format') or not value.has_canonical_format:
+            value = value.copy()
+            value.sum_duplicates()
+
         shape = cast(tuple[int, ...], value.shape)
         if len(shape) == 1:
             self._dim = shape[0]
@@ -115,10 +120,18 @@ class SparseVector:
 
         if hasattr(value, 'coords'):
             # scipy 1.13+
-            self._indices = value.coords[-1].tolist()
+            indices = value.coords[-1].tolist()
         else:
-            self._indices = value.col.tolist()
-        self._values = [float(v) for v in value.data]
+            indices = value.col.tolist()
+
+        elements = [(i, v) for i, v in zip(indices, value.data) if v != 0]
+
+        # has_canonical_format added in scipy 1.12+
+        if not hasattr(value, 'has_canonical_format') or not value.has_canonical_format:
+            elements.sort()
+
+        self._indices = [v[0] for v in elements]
+        self._values = [float(v[1]) for v in elements]
 
     def _from_dense(self, value: list[float] | ndarray, /) -> None:
         self._dim = len(value)

{pgvector-0.5.0 → pgvector-0.5.1}/pgvector.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pgvector
-Version: 0.5.0
+Version: 0.5.1
 Summary: pgvector support for Python
 Author-email: Andrew Kane <andrew@ankane.org>
 License-Expression: MIT

{pgvector-0.5.0 → pgvector-0.5.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "pgvector"
-version = "0.5.0"
+version = "0.5.1"
 description = "pgvector support for Python"
 readme = "README.md"
 authors = [

{pgvector-0.5.0 → pgvector-0.5.1}/tests/test_sparse_vector.py

@@ -54,12 +54,22 @@ class TestSparseVector:
         if np is None or sparse is None:
             pytest.skip('NumPy and SciPy required')
 
-        arr = sparse.coo_array(np.array([1, 0, 2, 0, 3, 0]))
+        arr = sparse.coo_array(([2, 3, 1, 0], ([2, 4, 0, 3],)), shape=(6,))
         vec = SparseVector(arr)
         assert vec.to_list() == [1, 0, 2, 0, 3, 0]
         assert vec.indices() == [0, 2, 4]
         assert isinstance(vec.values()[0], float)
 
+    def test_coo_array_duplicates(self) -> None:
+        if np is None or sparse is None:
+            pytest.skip('NumPy and SciPy required')
+
+        arr = sparse.coo_array(([1, 2], ([1, 1],)), shape=(3,))
+        assert arr.todense().tolist() == [0, 3, 0]
+        vec = SparseVector(arr)
+        assert vec.to_list() == [0, 3, 0]
+        assert vec.indices() == [1]
+
     def test_coo_array_dimensions(self) -> None:
         if np is None or sparse is None:
             pytest.skip('NumPy and SciPy required')