UPD: docstring/whitespace/blank line

mapie/control_risk/crc_rcps.py

@@ -3,7 +3,7 @@
 from typing import Optional, Tuple
 
 
-def _get_r_hat_plus(
+def get_r_hat_plus(
  risks: NDArray,
  lambdas: NDArray,
  method: Optional[str],
@@ -16,7 +16,6 @@ def _get_r_hat_plus(
 
  Parameters
  ----------
-
  risks: ArrayLike of shape (n_samples_cal, n_lambdas)
  The risk for each observation for each threshold
 
@@ -28,16 +27,15 @@ def _get_r_hat_plus(
  Correspond to the method use to control recall
  score. Could be either CRC or RCPS.
 
- bound: str
+ bound: Optional[str]
  Bounds to compute. Either hoeffding, bernstein or wsr.
 
- delta: float
+ delta: Optional[float]
  Level of confidence.
 
- sigma_init : float, optional
+ sigma_init : Optional[float]
  First variance in the sigma_hat array. The default
  value is the same as in the paper implementation [1].
- By default .25
 
  Returns
  -------
@@ -150,7 +148,7 @@ def _get_r_hat_plus(
  return r_hat, r_hat_plus
 
 
-def _find_lambda_star(
+def find_lambda_star(
  lambdas: NDArray,
  r_hat_plus: NDArray,
  alpha_np: NDArray

mapie/control_risk/ltt.py

@@ -16,11 +16,8 @@ def ltt_procedure(
 ) -> Tuple[List[List[Any]], NDArray]:
  """
  Apply the Learn-Then-Test procedure for risk control.
- This will apply learn then test procedure for
- risk control.
  Note that we will do a multiple test for ``r_hat`` that are
  less than level ``alpha_np``.
-
  The procedure follows the instructions in [1]:
  - Calculate p-values for each lambdas descretized
  - Apply a family wise error rate algorithm,

mapie/multi_label_classification.py

@@ -20,8 +20,8 @@
 
 from .control_risk.ltt import find_lambda_control_star, ltt_procedure
 from .control_risk.risks import compute_risk_precision, compute_risk_recall
-from .control_risk.crc_rcps import _get_r_hat_plus
-from .control_risk.crc_rcps import _find_lambda_star
+from .control_risk.crc_rcps import get_r_hat_plus
+from .control_risk.crc_rcps import find_lambda_star
 
 
 class MapieMultiLabelClassifier(BaseEstimator, ClassifierMixin):
@@ -122,6 +122,10 @@ class MapieMultiLabelClassifier(BaseEstimator, ClassifierMixin):
  [2] Angelopoulos, Anastasios N., Stephen, Bates, Adam, Fisch, Lihua,
  Lei, and Tal, Schuster. "Conformal Risk Control." (2022).
 
+ [3] Angelopoulos, A. N., Bates, S., Candès, E. J., Jordan,
+ M. I., & Lei, L. (2021). Learn then test:
+ "Calibrating predictive algorithms to achieve risk control".
+
  Examples
  --------
  >>> import numpy as np
@@ -152,7 +156,7 @@ class MapieMultiLabelClassifier(BaseEstimator, ClassifierMixin):
  "single_estimator_",
  "risks"
  ]
- sigma_init = 0.25
+ sigma_init = 0.25 # Value given in the paper [1]
  cal_size = .3
 
  def __init__(
@@ -680,11 +684,11 @@ def predict(
  )
 
  else:
- self.r_hat, self.r_hat_plus = _get_r_hat_plus(
+ self.r_hat, self.r_hat_plus = get_r_hat_plus(
  self.risks, self.lambdas, self.method,
  bound, delta, self.sigma_init
  )
- self.lambdas_star = _find_lambda_star(
+ self.lambdas_star = find_lambda_star(
  self.lambdas, self.r_hat_plus, alpha_np
  )
  y_pred_proba_array = (