add notebook on the cause of miscalibration

.github/workflows/testing.yml

@@ -28,6 +28,7 @@ jobs:
             pixi run check-generated-predictions
             pixi run build-calibration-curve
             pixi run different-calibration-curves
+            pixi run causes-miscalibration
 
       - name: Test if we can build the documentation
         if: matrix.os == 'ubuntu-latest'

book/_toc.yml

@@ -6,3 +6,4 @@ root: intro
 chapters:
 - file: content/notebooks/build_calibration_curve
 - file: content/notebooks/different_calibration_curves
+- file: content/notebooks/causes_miscalibration

content/python_files/causes_miscalibration.py

@@ -0,0 +1,61 @@
+# %%
+import numpy as np
+
+
+def xor_generator(n_samples=1_000, seed=0):
+    rng = np.random.default_rng(seed)
+    X = rng.uniform(low=-3, high=3, size=(n_samples, 2))
+    y = np.logical_xor(X[:, 0] > 0, X[:, 1] > 0)
+    return X, y
+
+
+# %%
+import matplotlib.pyplot as plt
+
+X, y = xor_generator(seed=0)
+_, ax = plt.subplots()
+ax.scatter(*X.T, c=y, cmap="coolwarm", alpha=0.5)
+ax.set(
+    xlim=(-3, 3),
+    ylim=(-3, 3),
+    xlabel="Feature 1",
+    ylabel="Feature 2",
+    title="XOR problem",
+    aspect="equal",
+)
+
+# %%
+from sklearn.preprocessing import SplineTransformer, PolynomialFeatures
+from sklearn.linear_model import LogisticRegression
+from sklearn.pipeline import make_pipeline
+
+model = make_pipeline(
+    SplineTransformer(n_knots=20),
+    PolynomialFeatures(degree=2, interaction_only=True),
+    LogisticRegression(),
+)
+model.fit(X, y)
+
+# %%
+from sklearn.inspection import DecisionBoundaryDisplay
+
+_, ax = plt.subplots()
+DecisionBoundaryDisplay.from_estimator(
+    model, X, ax=ax, cmap="coolwarm", response_method="predict_proba"
+)
+ax.scatter(*X.T, c=y, cmap="coolwarm", alpha=0.5)
+ax.set(
+    xlim=(-3, 3),
+    ylim=(-3, 3),
+    xlabel="Feature 1",
+    ylabel="Feature 2",
+    title="XOR problem",
+    aspect="equal",
+)
+
+# %%
+from sklearn.calibration import CalibrationDisplay
+
+CalibrationDisplay.from_estimator(model, X, y, strategy="quantile", n_bins=10)
+
+# %%

pixi.toml

@@ -11,6 +11,8 @@ generate-predictions = { cmd = "python _generate_predictions.py", cwd = "content
 build-calibration-curve = { cmd = "ipython build_calibration_curve.py", cwd = "content/python_files" }
 check-generated-predictions = { cmd = "python check_generated_predictions.py", cwd = "tests" }
 different-calibration-curves = { cmd = "ipython different_calibration_curves.py", cwd = "content/python_files" }
+causes-miscalibration = { cmd = "ipython causes_miscalibration.py", cwd = "content/python_files" }
+
 
 [dependencies]
 jupyterlab = ">=4.2.4,<5"