Create new conformity score

mapie/conformity_scores/conformity_scores.py

@@ -1,7 +1,7 @@
 from abc import ABCMeta, abstractmethod
 
 import numpy as np
-from typing import Tuple
+from typing import Tuple, Union
 
 from mapie._compatibility import np_nanquantile
 from mapie._typing import ArrayLike, NDArray
@@ -59,6 +59,7 @@ def get_signed_conformity_scores(
  X: ArrayLike,
  y: ArrayLike,
  y_pred: ArrayLike,
+ y_std: Union[ArrayLike, None],
  ) -> NDArray:
  """
  Placeholder for ``get_signed_conformity_scores``.
@@ -182,6 +183,7 @@ def get_conformity_scores(
  X: ArrayLike,
  y: ArrayLike,
  y_pred: ArrayLike,
+ y_std: Union[ArrayLike, None]
  ) -> NDArray:
  """
  Get the conformity score considering the symmetrical property if so.
@@ -202,7 +204,7 @@ def get_conformity_scores(
  NDArray of shape (n_samples,)
  Conformity scores.
  """
- conformity_scores = self.get_signed_conformity_scores(X, y, y_pred)
+ conformity_scores = self.get_signed_conformity_scores(X, y, y_pred, y_std)
  if self.consistency_check:
  self.check_consistency(X, y, y_pred, conformity_scores)
  if self.sym:

mapie/conformity_scores/residual_conformity_scores.py

@@ -588,7 +588,8 @@ def get_signed_conformity_scores(
  self,
  X: ArrayLike,
  y: ArrayLike,
- y_pred: ArrayLike
+ y_pred: ArrayLike,
+ y_std: Union[ArrayLike, None]
  ) -> NDArray:
  """
  Computes the signed conformity score = (y - y_pred) / r_pred.
@@ -666,3 +667,54 @@ def get_estimation_distribution(
  # import pdb; pdb.set_trace()
  print(conformity_scores)
  return np.add(y_pred, np.multiply(conformity_scores, r_pred.reshape(-1, 1)))
+
+
+
+class GPCrossConformityScore(ConformityScore):
+ """
+ Absolute conformity score.
+
+ The signed conformity score = y - y_pred.
+ The conformity score is symmetrical.
+
+ This is appropriate when the confidence interval is symmetrical and
+ its range is approximatively the same over the range of predicted values.
+ """
+
+ def __init__(
+ self,
+ ) -> None:
+ super().__init__(sym=True, consistency_check=True)
+
+ def get_signed_conformity_scores(
+ self,
+ X: ArrayLike,
+ y: ArrayLike,
+ y_pred: ArrayLike,
+ y_std: Union[ArrayLike, None]
+ ) -> NDArray:
+ """
+ Compute the signed conformity scores from the predicted values
+ and the observed ones, from the following formula:
+ signed conformity score = y - y_pred
+ """
+ y_std = np.maximum(self.eps, y_std)
+ return np.subtract(y, y_pred) / y_std
+
+
+ def get_estimation_distribution(
+ self,
+ X: ArrayLike,
+ y_pred: ArrayLike,
+ conformity_scores: ArrayLike
+ ) -> NDArray:
+ """
+ Compute samples of the estimation distribution from the predicted
+ values and the conformity scores, from the following formula:
+ signed conformity score = y - y_pred
+ <=> y = y_pred + signed conformity score
+
+ ``conformity_scores`` can be either the conformity scores or
+ the quantile of the conformity scores.
+ """
+ return np.add(y_pred, conformity_scores)

mapie/regression/test.py

@@ -2,6 +2,7 @@
 from sklearn.gaussian_process import GaussianProcessRegressor
 from sklearn.model_selection import train_test_split
 from mapie.regression import MapieRegressor
+from mapie.conformity_scores.residual_conformity_scores import GPCrossConformityScore
 
 def g(x):
  return (3 * x * np.sin(x) - 2 * x * np.cos(x) + ( x ** 3) / 40 - .5 * x ** 2 - 10 * x)
@@ -11,5 +12,5 @@ def g(x):
 X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2)
 gp = GaussianProcessRegressor().fit(X_train, y_train)
 
-mapie = MapieRegressor(estimator=gp, method="plus", cv=-1)
+mapie = MapieRegressor(estimator=gp, method="plus", cv=-1, conformity_score=GPCrossConformityScore(consistency_check=False))
 mapie.fit(X_train, y_train)