fix: Drop nptyping so numpy can move to 2.x

- nptyping capped numpy<2, pinning numpy 1.26.4 which has no
  Python 3.13 wheels, so Windows 3.13 CI built an experimental MinGW
  numpy that crashed with an access violation
- Replace nptyping annotations with numpy.typing.NDArray
- Require numpy>=2.1 (ships cp313 wheels) and relock
- Removes the np.bool8 deprecation warnings too

Author: georgedouzas

pyproject.toml

@@ -26,7 +26,7 @@ classifiers = [
     "Typing :: Typed",
 ]
 dependencies = [
-    "nptyping>=2.5.0",
+    "numpy>=2.1.0",
     "pandas>=2.3.3",
     "scikit-learn>=1.6.0",
     "cloudpickle>=2.0.0",

src/sportsbet/__init__.py

@@ -12,14 +12,14 @@
 
 import numpy as np
 import pandas as pd
-from nptyping import Bool, Float, Int, NDArray, Shape
+from numpy.typing import NDArray
 
 Param = dict[str, Any]
 ParamGrid = dict[str, list[Any]] | list[dict[str, list[Any]]]
 TrainData = tuple[pd.DataFrame, pd.DataFrame | None, pd.DataFrame]
 FixturesData = tuple[pd.DataFrame, None, pd.DataFrame]
-Data = NDArray[Shape['*, *'], Float]
-BoolData = NDArray[Shape['*, *'], Bool]
-Indices = NDArray[Shape['*, *'], Int]
+Data = NDArray[np.float64]
+BoolData = NDArray[np.bool_]
+Indices = NDArray[np.intp]
 Schema = list[tuple[str, type[int] | type[float] | type[object] | type[np.datetime64]]]
 OutputsMapping = dict[str, dict[str, Callable[..., pd.DataFrame]]]

src/sportsbet/evaluation/_base.py

@@ -12,7 +12,7 @@
 import cloudpickle
 import numpy as np
 import pandas as pd
-from nptyping import NDArray, Shape, String
+from numpy.typing import NDArray
 from sklearn.base import BaseEstimator, ClassifierMixin, MultiOutputMixin
 from sklearn.exceptions import NotFittedError
 from sklearn.utils import check_consistent_length, check_scalar
@@ -90,7 +90,7 @@ def __init__(
         self.init_cash = init_cash
         self.stake = stake
 
-    def _get_feature_names_odds(self: Self, O: pd.DataFrame) -> NDArray[Shape['*'], String]:  # noqa: F722
+    def _get_feature_names_odds(self: Self, O: pd.DataFrame) -> NDArray[np.str_]:
         # One odds column per selected market base, at the latest snapshot, ordered to match
         # `betting_markets_` so positional alignment with `Y` holds.
         columns = list(O.columns)

src/sportsbet/evaluation/_model_selection.py

@@ -12,7 +12,7 @@
 import numpy as np
 import pandas as pd
 from joblib import Parallel, delayed
-from nptyping import Float, NDArray, Shape
+from numpy.typing import NDArray
 from sklearn import get_config, set_config
 from sklearn.exceptions import NotFittedError
 from sklearn.model_selection import GridSearchCV, TimeSeriesSplit
@@ -399,7 +399,7 @@ def _scorer(
             estimator: BaseBettor,
             X: pd.DataFrame,
             Y: pd.DataFrame,
-            sample_weight: NDArray[Shape['*'], Float] | None = None,  # noqa: F722
+            sample_weight: NDArray[np.float64] | None = None,
             **kwargs: dict[str, Any],
         ) -> float:
             Y = Y[estimator.feature_names_out_]