Merge branch 'master' into 447-mapieregressor-sets-method-to-base-from-aci

Author: thibaultcordier

.bumpversion.cfg

@@ -16,6 +16,6 @@ search = version = "{current_version}"
 replace = version = "{new_version}"
 
 [bumpversion:file:CITATION.cff]
-search = version = "{current_version}"
-replace = version = "{new_version}"
+search = version: {current_version}
+replace = version: {new_version}
 

HISTORY.rst

@@ -2,9 +2,11 @@
 History
 =======
 
-0.8.3 (2024-**-**)
+0.8.4 (2024-**-**)
 ------------------
-* Fixed overloading of the value of the ‘method’ attribute when using MapieRegressor and MapieTimeSeriesRegressor
+
+* Fix the quantile formula to ensure valid coverage for any number of calibration data in `ConformityScore`.
+* Fix overloading of the value of the `method` attribute when using `MapieRegressor` and `MapieTimeSeriesRegressor`.
 * Fix conda versionning.
 * Reduce precision for test in `MapieCalibrator`.
 * Fix invalid certificate when downloading data.

README.rst

@@ -3,7 +3,7 @@
 |GitHubActions| |Codecov| |ReadTheDocs| |License| |PythonVersion| |PyPi| |Conda| |Release| |Commits| |DOI|
 
 .. |GitHubActions| image:: https://github.com/scikit-learn-contrib/MAPIE/actions/workflows/test.yml/badge.svg
-   :target: https://github.com/scikit-learn-contrib/MAPIE/actions
+    :target: https://github.com/scikit-learn-contrib/MAPIE/actions
 
 .. |Codecov| image:: https://codecov.io/gh/scikit-learn-contrib/MAPIE/branch/master/graph/badge.svg?token=F2S6KYH4V1
    :target: https://codecov.io/gh/scikit-learn-contrib/MAPIE
@@ -13,25 +13,25 @@
     :alt: Documentation Status
 
 .. |License| image:: https://img.shields.io/github/license/scikit-learn-contrib/MAPIE
-   :target: https://github.com/scikit-learn-contrib/MAPIE/blob/master/LICENSE
+    :target: https://github.com/scikit-learn-contrib/MAPIE/blob/master/LICENSE
 
 .. |PythonVersion| image:: https://img.shields.io/pypi/pyversions/mapie
-   :target: https://pypi.org/project/mapie/
+    :target: https://pypi.org/project/mapie/
 
 .. |PyPi| image:: https://img.shields.io/pypi/v/mapie
-   :target: https://pypi.org/project/mapie/
+    :target: https://pypi.org/project/mapie/
 
 .. |Conda| image:: https://img.shields.io/conda/vn/conda-forge/mapie
-   :target: https://anaconda.org/conda-forge/mapie
+    :target: https://anaconda.org/conda-forge/mapie
 
 .. |Release| image:: https://img.shields.io/github/v/release/scikit-learn-contrib/mapie
-   :target: https://github.com/scikit-learn-contrib/MAPIE/releases
+    :target: https://github.com/scikit-learn-contrib/MAPIE/releases
 
 .. |Commits| image:: https://img.shields.io/github/commits-since/scikit-learn-contrib/mapie/latest/master
-   :target: https://github.com/scikit-learn-contrib/MAPIE/commits/master
+    :target: https://github.com/scikit-learn-contrib/MAPIE/commits/master
 
 .. |DOI| image:: https://img.shields.io/badge/10.48550/arXiv.2207.12274-B31B1B.svg
-   :target: https://arxiv.org/abs/2207.12274
+    :target: https://arxiv.org/abs/2207.12274
 
 .. image:: https://github.com/scikit-learn-contrib/MAPIE/raw/master/doc/images/mapie_logo_nobg_cut.png
     :width: 400

mapie/conformity_scores/conformity_scores.py

@@ -214,7 +214,7 @@ def get_quantile(
         conformity_scores: NDArray,
         alpha_np: NDArray,
         axis: int,
-        method: str
+        reversed: bool = False
     ) -> NDArray:
         """
         Compute the alpha quantile of the conformity scores or the conformity
@@ -235,25 +235,32 @@ def get_quantile(
         axis: int
             The axis from which to compute the quantile.
 
-        method: str
-            ``"higher"`` or ``"lower"`` the method to compute the quantile.
+        reversed: bool
+            Boolean specifying whether we take the upper or lower quantile,
+            if False, the alpha quantile, otherwise the (1-alpha) quantile.
 
         Returns
         -------
         NDArray of shape (1, n_alpha) or (n_samples, n_alpha)
             The quantile of the conformity scores.
         """
-        n_ref = conformity_scores.shape[-1]
-        quantile = np.column_stack([
+        n_ref = conformity_scores.shape[1-axis]
+        n_calib = np.min(np.sum(~np.isnan(conformity_scores), axis=axis))
+        signed = 1-2*reversed
+
+        # Adapt alpha w.r.t upper/lower : alpha vs. 1-alpha
+        alpha_ref = (1-2*alpha_np)*reversed + alpha_np
+
+        # Adjust alpha w.r.t quantile correction
+        alpha_ref = np.ceil(alpha_ref*(n_calib+1))/n_calib
+
+        # Compute the target quantiles
+        quantile = signed * np.column_stack([
             np_nanquantile(
-                conformity_scores.astype(float),
-                _alpha,
-                axis=axis,
-                method=method
+                signed * conformity_scores, _alpha, axis=axis, method="lower"
             ) if 0 < _alpha < 1
-            else np.inf * np.ones(n_ref) if method == "higher"
-            else - np.inf * np.ones(n_ref)
-            for _alpha in alpha_np
+            else np.inf * np.ones(n_ref)
+            for _alpha in alpha_ref
         ])
         return quantile
 
@@ -281,7 +288,7 @@ def _beta_optimize(
         -------
         NDArray
             Array of betas minimizing the differences
-            ``(1-alpa+beta)-quantile - beta-quantile``.
+            ``(1-alpha+beta)-quantile - beta-quantile``.
         """
         beta_np = np.full(
             shape=(len(lower_bounds), len(alpha_np)),
@@ -405,26 +412,34 @@ def get_bounds(
                 X, y_pred_up, conformity_scores
             )
             bound_low = self.get_quantile(
-                conformity_scores_low, alpha_low, axis=1, method="lower"
+                conformity_scores_low, alpha_low, axis=1, reversed=True
             )
             bound_up = self.get_quantile(
-                conformity_scores_up, alpha_up, axis=1, method="higher"
+                conformity_scores_up, alpha_up, axis=1
             )
+
         else:
-            quantile_search = "higher" if self.sym else "lower"
-            alpha_low = 1 - alpha_np if self.sym else beta_np
-            alpha_up = 1 - alpha_np if self.sym else 1 - alpha_np + beta_np
+            if self.sym:
+                alpha_ref = 1-alpha_np
+                quantile_ref = self.get_quantile(
+                    conformity_scores[..., np.newaxis], alpha_ref, axis=0
+                )
+                quantile_low, quantile_up = -quantile_ref, quantile_ref
+
+            else:
+                alpha_low, alpha_up = beta_np, 1 - alpha_np + beta_np
+
+                quantile_low = self.get_quantile(
+                    conformity_scores[..., np.newaxis],
+                    alpha_low, axis=0, reversed=True
+                )
+                quantile_up = self.get_quantile(
+                    conformity_scores[..., np.newaxis],
+                    alpha_up, axis=0
+                )
 
-            quantile_low = self.get_quantile(
-                conformity_scores[..., np.newaxis],
-                alpha_low, axis=0, method=quantile_search
-            )
-            quantile_up = self.get_quantile(
-                conformity_scores[..., np.newaxis],
-                alpha_up, axis=0, method="higher"
-            )
             bound_low = self.get_estimation_distribution(
-                X, y_pred_low, signed * quantile_low
+                X, y_pred_low, quantile_low
             )
             bound_up = self.get_estimation_distribution(
                 X, y_pred_up, quantile_up

mapie/regression/regression.py

@@ -629,7 +629,7 @@ def predict(
 
             alpha_np = cast(NDArray, alpha)
             if not allow_infinite_bounds:
-                n = len(self.conformity_scores_)
+                n = np.sum(~np.isnan(self.conformity_scores_))
                 check_alpha_and_n_samples(alpha_np, n)
 
             y_pred, y_pred_low, y_pred_up = \

mapie/tests/test_conformity_scores.py

@@ -1,5 +1,3 @@
-from typing import Any
-
 import numpy as np
 import pytest
 from sklearn.linear_model import LinearRegression
@@ -382,18 +380,17 @@ def test_residual_normalised_prefit_get_estimation_distribution() -> None:
 @pytest.mark.parametrize("score", [AbsoluteConformityScore(),
                                    GammaConformityScore(),
                                    ResidualNormalisedScore()])
-@pytest.mark.parametrize("alpha", [[0.3], [0.5, 0.4]])
+@pytest.mark.parametrize("alpha", [[0.5], [0.5, 0.6]])
 def test_intervals_shape_with_every_score(
     score: ConformityScore,
-    alpha: Any
+    alpha: NDArray
 ) -> None:
+    estim = LinearRegression().fit(X_toy, y_toy)
     mapie_reg = MapieRegressor(
-        method="base", cv="split", conformity_score=score
+        estimator=estim, method="base", cv="prefit", conformity_score=score
     )
-    X = np.concatenate((X_toy, X_toy))
-    y = np.concatenate((y_toy, y_toy))
-    mapie_reg = mapie_reg.fit(X, y)
-    y_pred, intervals = mapie_reg.predict(X, alpha=alpha)
-    n_samples = X.shape[0]
+    mapie_reg = mapie_reg.fit(X_toy, y_toy)
+    y_pred, intervals = mapie_reg.predict(X_toy, alpha=alpha)
+    n_samples = X_toy.shape[0]
     assert y_pred.shape[0] == n_samples
     assert intervals.shape == (n_samples, 2, len(alpha))