chore: upgrade pyo3 to 0.23 (#188)

* chore: upgrade pyo3 to 0.23

Some of the APIs have changed due to changes in PyO3, but overall I
think this is simpler.

* Regenerate Python CI using latest maturin

* Update uv.lock

* Manually specify Python versions instead of using --find-interpreter

That flag is returning Python 3.13t (free-threadead) which isn't yet supported
by rust-numpy.

* Remove pypi interpreters from Windows builds

* Remove Python 3.13 interpreters from Windows builds

It looks like Python 3.13 isn't available wherever maturin-action
is running.

.github/workflows/python.yml

@@ -1,4 +1,4 @@
-# This file is autogenerated by maturin v1.6.0
+# This file is autogenerated by maturin v1.7.8
 # To update, run
 #
 #    maturin generate-ci github -m crates/pyaugurs/Cargo.toml
@@ -24,17 +24,17 @@ jobs:
     strategy:
       matrix:
         platform:
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: x86_64
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: x86
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: aarch64
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: armv7
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: s390x
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: ppc64le
     steps:
       - uses: actions/checkout@v4
@@ -45,7 +45,7 @@ jobs:
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist --find-interpreter --manifest-path crates/pyaugurs/Cargo.toml
+          args: --release --out dist -i 3.9 -i 3.10 -i 3.11 -i 3.12 -i 3.13 -i pypy-3.9 -i pypy-3.10 --manifest-path crates/pyaugurs/Cargo.toml
           sccache: 'true'
           manylinux: auto
       - name: Upload wheels
@@ -59,13 +59,13 @@ jobs:
     strategy:
       matrix:
         platform:
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: x86_64
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: x86
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: aarch64
-          - runner: ubuntu-latest
+          - runner: ubuntu-22.04
             target: armv7
     steps:
       - uses: actions/checkout@v4
@@ -76,7 +76,7 @@ jobs:
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist --find-interpreter --manifest-path crates/pyaugurs/Cargo.toml
+          args: --release --out dist -i 3.9 -i 3.10 -i 3.11 -i 3.12 -i 3.13 -i pypy-3.9 -i pypy-3.10 --manifest-path crates/pyaugurs/Cargo.toml
           sccache: 'true'
           manylinux: musllinux_1_2
       - name: Upload wheels
@@ -104,7 +104,7 @@ jobs:
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist --find-interpreter --manifest-path crates/pyaugurs/Cargo.toml
+          args: --release --out dist -i 3.9 -i 3.10 -i 3.11 -i 3.12 --manifest-path crates/pyaugurs/Cargo.toml
           sccache: 'true'
       - name: Upload wheels
         uses: actions/upload-artifact@v4
@@ -117,7 +117,7 @@ jobs:
     strategy:
       matrix:
         platform:
-          - runner: macos-12
+          - runner: macos-13
             target: x86_64
           - runner: macos-14
             target: aarch64
@@ -130,7 +130,7 @@ jobs:
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist --find-interpreter --manifest-path crates/pyaugurs/Cargo.toml
+          args: --release --out dist -i 3.9 -i 3.10 -i 3.11 -i 3.12 -i 3.13 -i pypy-3.9 -i pypy-3.10 --manifest-path crates/pyaugurs/Cargo.toml
           sccache: 'true'
       - name: Upload wheels
         uses: actions/upload-artifact@v4
@@ -139,7 +139,7 @@ jobs:
           path: dist
 
   sdist:
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
       - name: Build sdist
@@ -155,7 +155,7 @@ jobs:
 
   release:
     name: Release
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     if: "startsWith(github.ref, 'refs/tags/pyaugurs-')"
     needs: [linux, musllinux, windows, macos, sdist]
     permissions:

crates/pyaugurs/Cargo.toml

@@ -26,8 +26,8 @@ augurs-ets = { workspace = true, features = ["mstl"] }
 augurs-forecaster.workspace = true
 augurs-mstl.workspace = true
 augurs-seasons.workspace = true
-numpy = "0.21.0"
-pyo3 = { version = "0.21.0", features = ["extension-module"] }
+numpy = "0.23.0"
+pyo3 = { version = "0.23.3", features = ["extension-module"] }
 tracing = { version = "0.1.37", features = ["log"] }
 
 [lints]

crates/pyaugurs/augurs.pyi

@@ -48,15 +48,12 @@ class Forecast:
     def lower(self) -> npt.NDArray[np.float64] | None: ...
     def upper(self) -> npt.NDArray[np.float64] | None: ...
 
-class PyTrendModel:
-    def __init__(self, trend_model: TrendModel) -> None: ...
-
 class MSTL:
     @classmethod
     def ets(cls, periods: Sequence[int]) -> "MSTL": ...
     @classmethod
     def custom_trend(
-        cls, periods: Sequence[int], trend_model: PyTrendModel
+        cls, periods: Sequence[int], trend_model: TrendModel
     ) -> "MSTL": ...
     def fit(self, y: npt.NDArray[np.float64]) -> None: ...
     def predict(self, horizon: int, level: float | None) -> Forecast: ...

crates/pyaugurs/src/clustering.rs

@@ -84,10 +84,6 @@ impl Dbscan {
         distance_matrix: InputDistanceMatrix<'_>,
     ) -> PyResult<Py<PyArray1<isize>>> {
         let distance_matrix = distance_matrix.try_into()?;
-        Ok(self
-            .inner
-            .fit(&distance_matrix)
-            .into_pyarray_bound(py)
-            .into())
+        Ok(self.inner.fit(&distance_matrix).into_pyarray(py).into())
     }
 }

crates/pyaugurs/src/distance.rs

@@ -50,7 +50,7 @@ impl DistanceMatrix {
                 *elem = *val;
             }
         }
-        arr.into_pyarray_bound(py).into()
+        arr.into_pyarray(py).into()
     }
 }
 

crates/pyaugurs/src/dtw.rs

@@ -148,6 +148,7 @@ impl Dtw {
     }
 
     #[new]
+    #[pyo3(signature = (window=None, distance_fn=None, max_distance=None, lower_bound=None, upper_bound=None))]
     fn new(
         window: Option<usize>,
         distance_fn: Option<&str>,

crates/pyaugurs/src/ets.rs

@@ -63,6 +63,7 @@ impl AutoETS {
     /// # Errors
     ///
     /// This function will return an error if no model has been fit yet (using [`AutoETS::fit`]).
+    #[pyo3(signature = (horizon, level=None))]
     pub fn predict(&self, horizon: usize, level: Option<f64>) -> PyResult<Forecast> {
         self.fitted
             .as_ref()
@@ -80,6 +81,7 @@ impl AutoETS {
     /// # Errors
     ///
     /// This function will return an error if no model has been fit yet (using [`AutoETS::fit`]).
+    #[pyo3(signature = (level=None))]
     pub fn predict_in_sample(&self, level: Option<f64>) -> PyResult<Forecast> {
         self.fitted
             .as_ref()

crates/pyaugurs/src/lib.rs

@@ -38,6 +38,7 @@ impl From<Forecast> for augurs_core::Forecast {
 #[pymethods]
 impl Forecast {
     #[new]
+    #[pyo3(signature = (point, level=None, lower=None, upper=None))]
     fn new(
         py: Python<'_>,
         point: Py<PyArray1<f64>>,
@@ -80,23 +81,23 @@ impl Forecast {
         // We could also use `into_pyarray` to construct the
         // numpy arrays in the Rust heap; let's see which ends up being
         // faster and more convenient.
-        self.inner.point.to_pyarray_bound(py).into()
+        self.inner.point.to_pyarray(py).into()
     }
 
     /// Get the lower prediction interval.
     fn lower(&self, py: Python<'_>) -> Option<Py<PyArray1<f64>>> {
         self.inner
             .intervals
             .as_ref()
-            .map(|x| x.lower.to_pyarray_bound(py).into())
+            .map(|x| x.lower.to_pyarray(py).into())
     }
 
     /// Get the upper prediction interval.
     fn upper(&self, py: Python<'_>) -> Option<Py<PyArray1<f64>>> {
         self.inner
             .intervals
             .as_ref()
-            .map(|x| x.upper.to_pyarray_bound(py).into())
+            .map(|x| x.upper.to_pyarray(py).into())
     }
 }
 
@@ -106,7 +107,6 @@ fn augurs(_py: Python<'_>, m: &Bound<'_, PyModule>) -> PyResult<()> {
     // pyo3_log::init();
     m.add_class::<ets::AutoETS>()?;
     m.add_class::<mstl::MSTL>()?;
-    m.add_class::<trend::PyTrendModel>()?;
     m.add_class::<Forecast>()?;
     m.add_class::<clustering::Dbscan>()?;
     m.add_class::<distance::DistanceMatrix>()?;

crates/pyaugurs/src/mstl.rs

@@ -44,16 +44,30 @@ impl MSTL {
         }
     }
 
-    /// Create a new MSTL model with the given periods using provided trend model.
+    /// Create a new MSTL model with the given periods using the custom Python trend model.
+    ///
+    /// The custom trend model must implement the following methods:
+    ///
+    /// - `fit(self, y: np.ndarray) -> None`
+    /// - `predict(self, horizon: int, level: float | None = None) -> augurs.Forecast`
+    /// - `predict_in_sample(self, level: float | None = None) -> augurs.Forecast`
     #[classmethod]
     pub fn custom_trend(
         _cls: &Bound<'_, PyType>,
         periods: Vec<usize>,
-        trend_model: PyTrendModel,
+        trend_model: Py<PyAny>,
     ) -> Self {
-        let trend_model_name = trend_model.name().to_string();
+        let trend_model_name = Python::with_gil(|py| {
+            let trend_model = trend_model.bind(py).get_type();
+            trend_model
+                .name()
+                .map_or_else(|_| "unknown Python class".into(), |s| s.to_string())
+        });
         Self {
-            forecaster: Forecaster::new(MSTLModel::new(periods, Box::new(trend_model))),
+            forecaster: Forecaster::new(MSTLModel::new(
+                periods,
+                Box::new(PyTrendModel::new(trend_model)),
+            )),
             trend_model_name,
             fit: false,
         }
@@ -72,6 +86,7 @@ impl MSTL {
     /// intervals at the given level.
     ///
     /// If provided, `level` must be a float between 0 and 1.
+    #[pyo3(signature = (horizon, level=None))]
     pub fn predict(&self, horizon: usize, level: Option<f64>) -> PyResult<Forecast> {
         self.forecaster
             .predict(horizon, level)
@@ -83,6 +98,7 @@ impl MSTL {
     /// intervals at the given level.
     ///
     /// If provided, `level` must be a float between 0 and 1.
+    #[pyo3(signature = (level=None))]
     pub fn predict_in_sample(&self, level: Option<f64>) -> PyResult<Forecast> {
         self.forecaster
             .predict_in_sample(level)

crates/pyaugurs/src/seasons.rs

@@ -16,6 +16,7 @@ use augurs_seasons::{Detector, PeriodogramDetector};
 ///                   The default is 0.9.
 /// :return: an array of season lengths.
 #[pyfunction]
+#[pyo3(signature = (y, min_period=None, max_period=None, threshold=None))]
 pub fn seasonalities(
     py: Python<'_>,
     y: PyReadonlyArray1<'_, f64>,
@@ -35,9 +36,5 @@ pub fn seasonalities(
         builder = builder.threshold(threshold);
     }
 
-    Ok(builder
-        .build()
-        .detect(y.as_slice()?)
-        .to_pyarray_bound(py)
-        .into())
+    Ok(builder.build().detect(y.as_slice()?).to_pyarray(py).into())
 }

crates/pyaugurs/src/trend.rs

@@ -10,8 +10,9 @@
 //! - `fit(self, y: np.ndarray) -> None`
 //! - `predict(self, horizon: int, level: float | None = None) -> augurs.Forecast`
 //! - `predict_in_sample(self, level: float | None = None) -> augurs.Forecast`
+
 use numpy::ToPyArray;
-use pyo3::{exceptions::PyException, prelude::*};
+use pyo3::{exceptions::PyException, prelude::*, types::PyAnyMethods};
 
 use augurs_mstl::{FittedTrendModel, TrendModel};
 
@@ -28,8 +29,8 @@ use crate::Forecast;
 /// - `predict(self, horizon: int, level: float | None = None) -> augurs.Forecast`
 /// - `predict_in_sample(self, level: float | None = None) -> augurs.Forecast`
 #[pyclass(name = "TrendModel")]
-#[derive(Clone, Debug)]
-pub struct PyTrendModel {
+#[derive(Debug)]
+pub(crate) struct PyTrendModel {
     model: Py<PyAny>,
 }
 
@@ -44,7 +45,7 @@ impl PyTrendModel {
     /// The returned PyTrendModel can be used in MSTL models using the
     /// `custom_trend` method of the MSTL class.
     #[new]
-    pub fn new(model: Py<PyAny>) -> Self {
+    pub(crate) fn new(model: Py<PyAny>) -> Self {
         Self { model }
     }
 }
@@ -56,7 +57,7 @@ impl TrendModel for PyTrendModel {
                 .bind(py)
                 .get_type()
                 .name()
-                .map(|s| s.into_owned().into())
+                .map(|s| s.to_string().into())
         })
         .unwrap_or_else(|_| "unknown Python class".into())
     }
@@ -68,21 +69,18 @@ impl TrendModel for PyTrendModel {
         Box<dyn FittedTrendModel + Sync + Send>,
         Box<dyn std::error::Error + Send + Sync + 'static>,
     > {
-        // TODO - `fitted` should be a `PyFittedTrendModel`
-        // which should implement `Fit` and `FittedTrendModel`
-        Python::with_gil(|py| {
-            let np = y.to_pyarray_bound(py);
-            self.model.call_method1(py, "fit", (np,))
+        let model = Python::with_gil(|py| {
+            let np = y.to_pyarray(py);
+            self.model.call_method1(py, "fit", (np,))?;
+            Ok::<_, PyErr>(self.model.clone_ref(py))
         })?;
-        Ok(Box::new(PyFittedTrendModel {
-            model: self.model.clone(),
-        }) as _)
+        Ok(Box::new(PyFittedTrendModel { model }) as _)
     }
 }
 
 /// A wrapper for a Python trend model that has been fitted to data.
 #[derive(Debug)]
-pub struct PyFittedTrendModel {
+pub(crate) struct PyFittedTrendModel {
     model: Py<PyAny>,
 }