Merge pull request #184 from scikit-learn-contrib/fix_typing_cqr

Fix typing cqr

Author: vtaquet

HISTORY.rst

@@ -2,7 +2,7 @@
 History
 =======
 
-0.3.3 (2022-XX-XX)
+0.4.0 (2022-06-24)
 ------------------
 * Relax and fix typing
 * Add Split Conformal Quantile Regression

doc/images/quickstart_1.png

@@ -144,13 +144,7 @@ prediction intervals.
     for strategy, params in STRATEGIES.items():
         if strategy == "conformalized_quantile_regression":
             mapie = MapieQuantileRegressor(polyn_model_quant, **params)
-            X_train, X_calib, y_train, y_calib = train_test_split(
-                X_train,
-                y_train,
-                test_size=0.3,
-                random_state=1
-            )
-            mapie.fit(X_train, y_train, X_calib, y_calib)
+            mapie.fit(X_train, y_train, random_state=1)
             y_pred[strategy], y_pis[strategy] = mapie.predict(X_test)
         else:  
             mapie = MapieRegressor(polyn_model, **params)
@@ -462,13 +456,7 @@ prediction intervals.
     for strategy, params in STRATEGIES.items():
         if strategy == "conformalized_quantile_regression":
             mapie = MapieQuantileRegressor(polyn_model_quant, **params)
-            X_train, X_calib, y_train, y_calib = train_test_split(
-                X_train,
-                y_train,
-                test_size=0.3,
-                random_state=1
-            )
-            mapie.fit(X_train, y_train, X_calib, y_calib)
+            mapie.fit(X_train, y_train, random_state=1)
             y_pred[strategy], y_pis[strategy] = mapie.predict(X_test)
         else:  
             mapie = MapieRegressor(polyn_model, **params)
@@ -567,7 +555,7 @@ next figure.
         recap ={}
         for i in range(len(bins)-1):
             bin1, bin2 = bins[i], bins[i+1]
-            name = f"{bin1} to {bin2}"
+            name = f"[{bin1}, {bin2}]"
             recap[name] = []
             for strategy in STRATEGIES:
                 indices = np.where((X_test>=bins[i])*(X_test<=bins[i+1]))
@@ -590,6 +578,9 @@ next figure.
     heteroscedastic_coverage.T.plot.bar(figsize=(12, 4), alpha=0.7)
     plt.axhline(0.95, ls="--", color="k")
     plt.ylabel("Conditional coverage")
+    plt.xlabel("x bins")
+    plt.xticks(rotation=0)
+    plt.ylim(0.8, 1.0)
     plt.legend(loc=[1, 0])
 
 
@@ -790,13 +781,7 @@ strategies.
     for strategy, params in STRATEGIES.items():
         if strategy == "conformalized_quantile_regression":
             mapie = MapieQuantileRegressor(polyn_model_quant, **params)
-            X_train, X_calib, y_train, y_calib = train_test_split(
-                X_train,
-                y_train,
-                test_size=0.3,
-                random_state=1
-            )
-            mapie.fit(X_train, y_train, X_calib, y_calib)
+            mapie.fit(X_train, y_train, random_state=1)
             y_pred[strategy], y_pis[strategy] = mapie.predict(X_test)
         else:  
             mapie = MapieRegressor(polyn_model, **params)
@@ -1068,7 +1053,6 @@ method and compare their prediction interval.
 
 .. parsed-literal::
 
-    Metal device set to: Apple M1
 
 
 .. code-block:: python

doc/tutorial_regression.rst

@@ -15,7 +15,6 @@
 import scipy
 import numpy as np
 from sklearn.linear_model import LinearRegression, QuantileRegressor
-from sklearn.model_selection import train_test_split
 from sklearn.pipeline import Pipeline
 from sklearn.preprocessing import PolynomialFeatures
 from matplotlib import pyplot as plt
@@ -25,6 +24,8 @@
 from mapie.subsample import Subsample
 from mapie._typing import NDArray
 
+random_state = 42
+
 
 def f(x: NDArray) -> NDArray:
     """Polynomial function used to generate one-dimensional data"""
@@ -59,7 +60,7 @@ def get_heteroscedastic_data(
         [3]: y_true
         [4]: y_true_sigma
     """
-    np.random.seed(59)
+    np.random.seed(random_state)
     q95 = scipy.stats.norm.ppf(0.95)
     X_train = np.linspace(0, 1, n_train)
     X_true = np.linspace(0, 1, n_true)
@@ -158,22 +159,8 @@ def plot_1d_data(
             polyn_model_quant,
             **params
         )
-        X_train_split, X_calib, y_train_spit, y_calib = train_test_split(
-            X_train,
-            y_train,
-            test_size=0.3,
-            random_state=1
-        )
-        mapie.fit(
-            X_train_split.reshape(-1, 1),
-            y_train_spit,
-            X_calib.reshape(-1, 1),
-            y_calib
-        )
-        y_pred, y_pis = mapie.predict(
-            X_test.reshape(-1, 1)
-        )
-        X_train, y_train = X_train_split, y_train_spit
+        mapie.fit(X_train.reshape(-1, 1), y_train, random_state=random_state)
+        y_pred, y_pis = mapie.predict(X_test.reshape(-1, 1))
     else:
         mapie = MapieRegressor(  # type: ignore
             polyn_model,