update to latest mypy and exclude Python 3.6

azure-pipelines.yml

@@ -63,7 +63,7 @@ jobs:
  - script: |
  python -m mypy thinc
  displayName: 'Run mypy'
- condition: ne(variables['python.version'], '3.10')
+ condition: ne(variables['python.version'], '3.6')
 
  - task: DeleteFiles@1
  inputs:

requirements.txt

@@ -23,7 +23,7 @@ pytest-cov>=2.7.0,<2.8.0
 coverage>=5.0.0,<6.0.0
 mock>=2.0.0,<3.0.0
 flake8>=3.5.0,<3.6.0
-mypy>=0.901,<0.970; platform_machine!='aarch64'
+mypy>=0.980,<0.990; platform_machine != "aarch64" and python_version >= "3.7"
 types-mock>=0.1.1
 types-contextvars>=0.1.2; python_version < "3.7"
 types-dataclasses>=0.1.3; python_version < "3.7"

thinc/backends/ops.py

@@ -373,7 +373,7 @@ def pad( # noqa: F811
  # array sizes.
  length = (length + (round_to - 1)) // round_to * round_to
  final_shape = (len(seqs), length) + seqs[0].shape[1:]
- output: Array3d = self.alloc(final_shape, dtype=seqs[0].dtype)
+ output: Array3d = cast(Array3d, self.alloc(final_shape, dtype=seqs[0].dtype))
  for i, arr in enumerate(seqs):
  # It's difficult to convince this that the dtypes will match.
  output[i, : arr.shape[0]] = arr # type: ignore[assignment, call-overload]
@@ -451,7 +451,7 @@ def get_dropout_mask(self, shape: Shape, drop: Optional[float]) -> FloatsXd:
  if drop is None or drop <= 0:
  return self.xp.ones(shape, dtype="f")
  elif drop >= 1.0:
- return self.alloc(shape)
+ return self.alloc_f(shape)
  coinflips = self.xp.random.uniform(0.0, 1.0, shape)
  mask = (coinflips >= drop) / (1.0 - drop)
  return cast(FloatsXd, self.asarray(mask, dtype="float32"))
@@ -463,7 +463,7 @@ def alloc1f(
  dtype: Optional[DTypesFloat] = "float32",
  zeros: bool = True,
  ) -> Floats1d:
- return self.alloc((d0,), dtype=dtype, zeros=zeros)
+ return cast(Floats1d, self.alloc((d0,), dtype=dtype, zeros=zeros))
 
  def alloc2f(
  self,
@@ -473,7 +473,7 @@ def alloc2f(
  dtype: Optional[DTypesFloat] = "float32",
  zeros: bool = True,
  ) -> Floats2d:
- return self.alloc((d0, d1), dtype=dtype, zeros=zeros)
+ return cast(Floats2d, self.alloc((d0, d1), dtype=dtype, zeros=zeros))
 
  def alloc3f(
  self,
@@ -484,7 +484,7 @@ def alloc3f(
  dtype: Optional[DTypesFloat] = "float32",
  zeros: bool = True,
  ) -> Floats3d:
- return self.alloc((d0, d1, d2), dtype=dtype, zeros=zeros)
+ return cast(Floats3d, self.alloc((d0, d1, d2), dtype=dtype, zeros=zeros))
 
  def alloc4f(
  self,
@@ -496,7 +496,7 @@ def alloc4f(
  dtype: Optional[DTypesFloat] = "float32",
  zeros: bool = True,
  ) -> Floats4d:
- return self.alloc((d0, d1, d2, d3), dtype=dtype, zeros=zeros)
+ return cast(Floats4d, self.alloc((d0, d1, d2, d3), dtype=dtype, zeros=zeros))
 
  def alloc_f(
  self,
@@ -505,7 +505,7 @@ def alloc_f(
  dtype: Optional[DTypesFloat] = "float32",
  zeros: bool = True,
  ) -> FloatsXd:
- return self.alloc(shape, dtype=dtype, zeros=zeros)
+ return cast(FloatsXd, self.alloc(shape, dtype=dtype, zeros=zeros))
 
  def alloc1i(
  self,
@@ -514,7 +514,7 @@ def alloc1i(
  dtype: Optional[DTypesInt] = "int32",
  zeros: bool = True,
  ) -> Ints1d:
- return self.alloc((d0,), dtype=dtype, zeros=zeros)
+ return cast(Ints1d, self.alloc((d0,), dtype=dtype, zeros=zeros))
 
  def alloc2i(
  self,
@@ -524,7 +524,7 @@ def alloc2i(
  dtype: Optional[DTypesInt] = "int32",
  zeros: bool = True,
  ) -> Ints2d:
- return self.alloc((d0, d1), dtype=dtype, zeros=zeros)
+ return cast(Ints2d, self.alloc((d0, d1), dtype=dtype, zeros=zeros))
 
  def alloc3i(
  self,
@@ -535,7 +535,7 @@ def alloc3i(
  dtype: Optional[DTypesInt] = "int32",
  zeros: bool = True,
  ) -> Ints3d:
- return self.alloc((d0, d1, d2), dtype=dtype, zeros=zeros)
+ return cast(Ints3d, self.alloc((d0, d1, d2), dtype=dtype, zeros=zeros))
 
  def alloc4i(
  self,
@@ -547,7 +547,7 @@ def alloc4i(
  dtype: Optional[DTypesInt] = "int32",
  zeros: bool = True,
  ) -> Ints4d:
- return self.alloc((d0, d1, d2, d3), dtype=dtype, zeros=zeros)
+ return cast(Ints4d, self.alloc((d0, d1, d2, d3), dtype=dtype, zeros=zeros))
 
  def alloc_i(
  self,
@@ -556,15 +556,15 @@ def alloc_i(
  dtype: Optional[DTypesInt] = "int32",
  zeros: bool = True,
  ) -> IntsXd:
- return self.alloc(shape, dtype=dtype, zeros=zeros)
+ return cast(IntsXd, self.alloc(shape, dtype=dtype, zeros=zeros))
 
  def alloc(
  self,
  shape: Shape,
  *,
  dtype: Optional[DTypes] = "float32",
  zeros: bool = True,
- ) -> ArrayT:
+ ) -> ArrayXd:
  """Allocate an array of a certain shape."""
  if isinstance(shape, int):
  shape = (shape,)

thinc/initializers.py

@@ -75,7 +75,7 @@ def configure_glorot_uniform_init() -> Callable[[Shape], FloatsXd]:
 
 
 def zero_init(ops: Ops, shape: Shape) -> FloatsXd:
- return ops.alloc(shape)
+ return ops.alloc_f(shape)
 
 
 @registry.initializers("zero_init.v1")

thinc/layers/hashembed.py

@@ -62,7 +62,7 @@ def forward(
  nV = vectors.shape[0]
  nO = vectors.shape[1]
  if len(ids) == 0:
- output: Floats2d = model.ops.alloc((0, nO), dtype=vectors.dtype)
+ output: Floats2d = model.ops.alloc2f(0, nO, dtype=vectors.dtype)
  else:
  ids = model.ops.as_contig(ids, dtype="uint64")
  nN = ids.shape[0]

thinc/tests/layers/test_mxnet_wrapper.py

@@ -34,7 +34,7 @@ def answer() -> int:
 @pytest.fixture
 def X(input_size: int) -> Array2d:
  ops: Ops = get_current_ops()
- return ops.alloc(shape=(1, input_size))
+ return cast(Array2d, ops.alloc(shape=(1, input_size)))
 
 
 @pytest.fixture

thinc/types.py

@@ -162,7 +162,7 @@ def __bytes__(self) -> bytes: ...
  def __str__(self) -> str: ...
  def __repr__(self) -> str: ...
  def __copy__(self, order: str = ...): ...
- def __deepcopy__(self, memo: dict) -> ArrayT: ...
+ def __deepcopy__(self: SelfT, memo: dict) -> SelfT: ...
  def __lt__(self, other): ...
  def __le__(self, other): ...
  def __eq__(self, other): ...
@@ -224,7 +224,7 @@ def clip(self, a_min: Any, a_max: Any, out: Optional[ArrayT]) -> ArrayT: ...
  def max(self, axis: int = -1, out: Optional[ArrayT] = None) -> ArrayT: ...
  # def mean(self, axis: int = -1, dtype: Optional[DTypes] = None, out: Optional[SelfT] = None, keepdims: bool = False) -> "Array": ...
  def min(self, axis: int = -1, out: Optional[ArrayT] = None) -> ArrayT: ...
- def nonzero(self) -> ArrayT: ...
+ def nonzero(self: SelfT) -> SelfT: ...
  def prod(self, axis: int = -1, dtype: Optional[DTypes] = None, out: Optional[ArrayT] = None, keepdims: bool = False) -> ArrayT: ...
  def round(self, decimals: int = 0, out: Optional[ArrayT] = None) -> ArrayT: ...
  # def sum(self, axis: int = -1, dtype: Optional[DTypes] = None, out: Optional[ArrayT] = None, keepdims: bool = False) -> ArrayT: ...