Make type hints compatible with numpy 1.20.

Author: greschd

mypy.ini

@@ -31,9 +31,6 @@ strict_equality = True
 [mypy-setuptools.*]
 ignore_missing_imports = True
 
-[mypy-numpy.*]
-ignore_missing_imports = True
-
 [mypy-scipy.*]
 ignore_missing_imports = True
 

pytest.ini

@@ -2,7 +2,4 @@
 filterwarnings =
     error
     ignore::ImportWarning
-    ignore:numpy.dtype size changed
-    ignore:numpy.ufunc size changed
-    ignore:the imp module is deprecated
 junit_family = xunit2

setup.cfg

@@ -33,7 +33,7 @@ setup_requires =
 install_requires =
     numpy
     scipy>=0.15
-    h5py>=3.0.0
+    h5py>=3.2.0
     fsc.export
     symmetry-representation>=0.2
     click>=7.0, !=7.1.0

tbmodels/_tb_model.py

@@ -633,13 +633,15 @@ def from_wannier_files(  # pylint: disable=too-many-locals
                 wannier_pos_cartesian = np.array(wannier_pos_list_cartesian)
                 atom_pos_cartesian = np.array([a.pos for a in atom_list_cartesian])
                 if pos_kind == "wannier":
-                    pos_cartesian = wannier_pos_cartesian
+                    pos_cartesian: ty.Union[
+                        ty.List[np.ndarray], np.ndarray
+                    ] = wannier_pos_cartesian
                 elif pos_kind == "nearest_atom":
                     if distance_ratio_threshold < 1:
                         raise ValueError(
                             "Invalid value for 'distance_ratio_threshold': must be >= 1."
                         )
-                    pos_cartesian = []
+                    pos_cartesian = ty.cast(ty.List[np.ndarray], [])
                     for p in wannier_pos_cartesian:
                         p_reduced = la.solve(kwargs["uc"].T, np.array(p).T).T
                         T_base = np.floor(p_reduced)
@@ -1130,7 +1132,7 @@ def hamilton(
             H = pos_exponential.conjugate().transpose((0, 2, 1)) * H * pos_exponential
 
         if single_point:
-            return H[0]
+            return ty.cast(np.ndarray, H[0])
         return H
 
     def eigenval(
@@ -1149,7 +1151,7 @@ def eigenval(
         hamiltonians = self.hamilton(k)
         if hamiltonians.ndim == 3:
             return [la.eigvalsh(ham) for ham in hamiltonians]
-        return la.eigvalsh(hamiltonians)
+        return ty.cast(np.ndarray, la.eigvalsh(hamiltonians))
 
     # -------------------MODIFYING THE MODEL ----------------------------#
     def add_hop(
@@ -1320,7 +1322,7 @@ def set_sparse(self, sparse: bool = True):
         # change existing matrices
         with contextlib.suppress(AttributeError):
             for k, v in self.hop.items():
-                self.hop[k] = self._matrix_type(v)
+                self.hop[k] = self._matrix_type(v)  # type: ignore
 
     # If Python 3.4 support is dropped this could be made more straightforwardly
     # However, for now the default pickle protocol (and thus multiprocessing)
@@ -1592,7 +1594,7 @@ def change_unit_cell(  # pylint: disable=too-many-branches
                 )
             # convert to reduced coordinates
             if uc is None:
-                new_uc = self.uc
+                new_uc: ty.Optional[np.ndarray] = self.uc
                 uc_reduced = np.eye(self.dim)
             else:
                 new_uc = np.array(uc)

tbmodels/helpers.py

@@ -17,7 +17,7 @@ def matrix_to_hop(
     orbitals: ty.Optional[ty.Sequence[int]] = None,
     R: ty.Sequence[int] = (0, 0, 0),
     multiplier: float = 1.0,
-) -> ty.List[ty.List[ty.Union[complex, int, np.ndarray]]]:
+) -> ty.List[ty.Tuple[complex, int, int, np.ndarray]]:
     r"""
     Turns a square matrix into a series of hopping terms.
 
@@ -40,11 +40,11 @@ def matrix_to_hop(
     for i, row in enumerate(mat):
         for j, x in enumerate(row):
             hop.append(
-                [
+                (
                     multiplier * x,
                     orbitals[i],
                     orbitals[j],
                     np.array(R, dtype=int),
-                ]
+                )
             )
     return hop

tbmodels/kdotp.py

@@ -68,13 +68,16 @@ def hamilton(
         else:
             single_point = False
 
-        ham: np.ndarray = sum(
-            np.prod(k_array ** k_powers, axis=-1).reshape(-1, 1, 1)
-            * mat[np.newaxis, :, :]
-            for k_powers, mat in self.taylor_coefficients.items()
+        ham = ty.cast(
+            np.ndarray,
+            sum(
+                np.prod(k_array ** k_powers, axis=-1).reshape(-1, 1, 1)
+                * mat[np.newaxis, :, :]
+                for k_powers, mat in self.taylor_coefficients.items()
+            ),
         )
         if single_point:
-            return ham[0]
+            return ty.cast(np.ndarray, ham[0])
         return ham
 
     def eigenval(
@@ -93,4 +96,4 @@ def eigenval(
         hamiltonians = self.hamilton(k)
         if hamiltonians.ndim == 3:
             return [la.eigvalsh(ham) for ham in hamiltonians]
-        return la.eigvalsh(hamiltonians)
+        return ty.cast(np.ndarray, la.eigvalsh(hamiltonians))