fixed some misnomers

cyclic_boosting/generic_loss.py

@@ -13,7 +13,7 @@
 
 from cyclic_boosting.base import CyclicBoostingBase, gaussian_matching_by_quantiles, Feature, CBLinkPredictionsFactors
 from cyclic_boosting.link import LogLinkMixin, IdentityLinkMixin, LogitLinkMixin
-from cyclic_boosting.utils import continuous_cdf_from_discrete_pdf, get_X_column
+from cyclic_boosting.utils import continuous_quantile_from_discrete_pdf, get_X_column
 from cyclic_boosting.classification import get_beta_priors
 
 from typing import Tuple, Union
@@ -408,7 +408,7 @@ def quantile_global_scale(
     if weights is None:
         raise RuntimeError("The weights have to be initialized.")
 
-    global_scale_link_ = link_func(continuous_cdf_from_discrete_pdf(y, quantile))
+    global_scale_link_ = link_func(continuous_quantile_from_discrete_pdf(y, quantile))
 
     prior_pred_link_offset_ = None
     if prior_prediction_column is not None:

cyclic_boosting/quantile_matching.py

@@ -6,24 +6,24 @@
 from typing import Optional
 
 
-def cdf_fit_gaussian(quantiles: np.ndarray, cdf_values: np.ndarray, mode: Optional[str] = "ppf") -> callable:
+def quantile_fit_gaussian(quantiles: np.ndarray, quantile_values: np.ndarray, mode: Optional[str] = "ppf") -> callable:
     """
-    Interpolation between pairs of CDF values (potentially estimated by means
-    of quantile regression) and corresponding quantiles according to a
-    Gaussian distribution as assumed PDF.
+    Interpolation of a quantile function (with quantiles estimated, e.g., by
+    means of quantile regression) according to a Gaussian distribution as
+    assumed PDF.
 
     Parameters
     ----------
     quantiles : np.ndarray
-        quantile values
-    cdf_values : np.ndarray
-        CDF values corresponding to quantile values
+        quantiles (x values of quantile function)
+    quantile_values : np.ndarray
+        quantile values (y values of quantile function)
     mode : str
         decides about kind of returned callable, possible values are:
 
-            - ``ppf``: input quantile, output CDF value (default)
-            - ``dist``: fitted Gaussian (scipy function)
-            - ``cdf``: input CDF value, output quantile
+            - ``ppf``: quantile (default)
+            - ``dist``: fitted negative binomial (scipy function)
+            - ``cdf``: CDF function
 
     Returns
     -------
@@ -32,9 +32,9 @@ def cdf_fit_gaussian(quantiles: np.ndarray, cdf_values: np.ndarray, mode: Option
     """
 
     def f(x, mu, sigma):
-        return norm(loc=mu, scale=sigma).cdf(x)
+        return norm(loc=mu, scale=sigma).ppf(x)
 
-    mu, sigma = curve_fit(f, cdf_values, quantiles)[0]
+    mu, sigma = curve_fit(f, quantiles, quantile_values)[0]
     if mode == "ppf":
         return norm(mu, sigma).ppf
     elif mode == "dist":
@@ -45,25 +45,24 @@ def f(x, mu, sigma):
         raise Exception("Invalid mode.")
 
 
-def cdf_fit_gamma(quantiles: np.ndarray, cdf_values: np.ndarray, mode: Optional[str] = "ppf") -> callable:
+def quantile_fit_gamma(quantiles: np.ndarray, quantile_values: np.ndarray, mode: Optional[str] = "ppf") -> callable:
     """
-    Interpolation between pairs of CDF values (potentially estimated by means
-    of quantile regression) and corresponding quantiles according to a
-    Gamma distribution as assumed PDF (i.e., continuous, non-negative target
-    values).
+    Interpolation of a quantile function (with quantiles estimated, e.g., by
+    means of quantile regression) according to a Gamma distribution as assumed
+    PDF (i.e., continuous, non-negative target values).
 
     Parameters
     ----------
     quantiles : np.ndarray
-        quantile values
-    cdf_values : np.ndarray
-        CDF values corresponding to quantile values
+        quantiles (x values of quantile function)
+    quantile_values : np.ndarray
+        quantile values (y values of quantile function)
     mode : str
         decides about kind of returned callable, possible values are:
 
-            - ``ppf``: input quantile, output CDF value (default)
-            - ``dist``: fitted Gamma (scipy function)
-            - ``cdf``: input CDF value, output quantile
+            - ``ppf``: quantile (default)
+            - ``dist``: fitted negative binomial (scipy function)
+            - ``cdf``: CDF function
 
     Returns
     -------
@@ -72,9 +71,9 @@ def cdf_fit_gamma(quantiles: np.ndarray, cdf_values: np.ndarray, mode: Optional[
     """
 
     def f(x, alpha, beta):
-        return gamma(alpha, scale=1 / beta).cdf(x)
+        return gamma(alpha, scale=1 / beta).ppf(x)
 
-    alpha, beta = curve_fit(f, cdf_values, quantiles, p0=[2.0, 0.9])[0]
+    alpha, beta = curve_fit(f, quantiles, quantile_values, p0=[2.0, 0.9])[0]
     if mode == "ppf":
         return gamma(alpha, scale=1 / beta).ppf
     elif mode == "dist":
@@ -88,47 +87,52 @@ def f(x, alpha, beta):
 def _nbinom_cdf_mu_var(x: float, mu: float, var: float) -> callable:
     """
     Calculation of negative binomial parameters n and p from given mean and
-    variance, and subsequent call of cumulative distribution function.
+    variance, and subsequent call of its cumulative distribution function.
+
     Parameters
     ----------
     x : float
-        value of random variable following a negative binomial distribution
+        value of random variable following negative binomial distribution
     mu : float
         mean of negative binomial distribution
     var : float
         variance of negative binomial distribution
+
+    Returns
+    -------
+    callable
+        negative binomial cumulative distribution function
     """
     n = mu * mu / (var - mu)
     p = mu / var
     return nbinom(n, p).cdf(x)
 
 
-def cdf_fit_nbinom(quantiles: np.ndarray, cdf_values: np.ndarray, mode: Optional[str] = "ppf") -> callable:
+def quantile_fit_nbinom(quantiles: np.ndarray, quantile_values: np.ndarray, mode: Optional[str] = "ppf") -> callable:
     """
-    Interpolation between pairs of CDF values (potentially estimated by means
-    of quantile regression) and corresponding quantiles according to a
-    negative binomial distribution as assumed PDF (i.e., discrete, non-negative
-    target values).
+    Interpolation of a quantile function (with quantiles estimated, e.g., by
+    means of quantile regression) according to a negative binomial distribution
+    as assumed PDF (i.e., discrete, non-negative target values).
 
     Parameters
     ----------
     quantiles : np.ndarray
-        quantile values
-    cdf_values : np.ndarray
-        CDF values corresponding to quantile values
+        quantiles (x values of quantile function)
+    quantile_values : np.ndarray
+        quantile values (y values of quantile function)
     mode : str
         decides about kind of returned callable, possible values are:
 
-            - ``ppf``: input quantile, output CDF value (default)
+            - ``ppf``: quantile (default)
             - ``dist``: fitted negative binomial (scipy function)
-            - ``cdf``: input CDF value, output quantile
+            - ``cdf``: CDF function
 
     Returns
     -------
     callable
         fitted negative binomial function (see mode)
     """
-    mu, var = curve_fit(_nbinom_cdf_mu_var, cdf_values, quantiles, p0=[1.2, 1.4])[0]
+    mu, var = curve_fit(_nbinom_cdf_mu_var, quantile_values, quantiles, p0=[2.2, 2.4])[0]
     n = mu * mu / (var - mu)
     p = mu / var
     if mode == "ppf":
@@ -141,23 +145,23 @@ def cdf_fit_nbinom(quantiles: np.ndarray, cdf_values: np.ndarray, mode: Optional
         raise Exception("Invalid mode.")
 
 
-def cdf_fit_spline(quantiles: np.ndarray, cdf_values: np.ndarray) -> callable:
+def quantile_fit_spline(quantiles: np.ndarray, quantile_values: np.ndarray) -> callable:
     """
-    Interpolation between pairs of CDF values (potentially estimated by means
-    of quantile regression) and corresponding quantiles according to a
-    smoothing spline (i.e., arbitrary target distribution).
+    Interpolation of a quantile function (with quantiles estimated, e.g., by
+    means of quantile regression) according to a smoothing spline (i.e.,
+    arbitrary target distribution).
 
     Parameters
     ----------
     quantiles : np.ndarray
-        quantile values
-    cdf_values : np.ndarray
-        CDF values corresponding to quantile values
+        quantiles (x values of quantile function)
+    quantile_values : np.ndarray
+        quantile values (y values of quantile function)
 
     Returns
     -------
     callable
-        fitted spline function (input quantile, output CDF value)
+        spline fitted to quantile function
     """
-    spl = InterpolatedUnivariateSpline(quantiles, cdf_values, k=3, bbox=[0, 1], ext=3)
+    spl = InterpolatedUnivariateSpline(quantiles, quantile_values, k=3, bbox=[0, 1], ext=3)
     return spl

cyclic_boosting/utils.py

@@ -981,10 +981,10 @@ def get_feature_column_names(X, exclude_columns=[]):
     return features
 
 
-def continuous_cdf_from_discrete_pdf(y, quantile):
+def continuous_quantile_from_discrete_pdf(y, quantile):
     """
-    Calculates a continous CDF value approximation for a given quantile from an
-    array of potentially discrete values.
+    Calculates a continous quantile value approximation for a given quantile
+    from an array of potentially discrete values.
 
     Parameters
     ----------
@@ -996,7 +996,7 @@ def continuous_cdf_from_discrete_pdf(y, quantile):
     Returns
     -------
     float
-        calculated CDF value
+        calculated quantile value
     """
     sorted_y = np.sort(y)
     quantile_index = int(quantile * (len(y) - 1))
@@ -1032,7 +1032,7 @@ def get_normalized_values(values: Iterable) -> List[float]:
 def smear_discrete_cdftruth(cdf_func: callable, y: int) -> float:
     """
     Smearing of the CDF value of a sample from a discrete random variable. Main
-    usage is for a histogram of CDF values to check estimated individual
+    usage is for a histogram of CDF values to check an estimated individual
     probability distribution (should be flat).
 
     Parameters

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "cyclic-boosting"
-version = "1.1.1"
+version = "1.1.2"
 description = "Implementation of Cyclic Boosting machine learning algorithms"
 authors = ["Blue Yonder GmbH"]
 packages = [{include = "cyclic_boosting"}]