Merge pull request #204 from guillermo-navas-palencia/develop

Develop

README.rst

@@ -74,7 +74,7 @@ OptBinning requires
 * numpy (>=1.16.1)
 * ortools (>=7.2)
 * pandas
-* ropwr (>=0.2.0)
+* ropwr (>=0.4.0)
 * scikit-learn (>=0.22.0)
 * scipy (>=1.6.0)
 

doc/source/conf.py

@@ -22,9 +22,9 @@
 author = 'Guillermo Navas-Palencia'
 
 # The short X.Y version
-version = '0.16.0'
+version = '0.16.1'
 # The full version, including alpha/beta/rc tags
-release = '0.16.0'
+release = '0.16.1'
 
 
 # -- General configuration ---------------------------------------------------

doc/source/outlier.rst

@@ -17,3 +17,9 @@ Outlier detection
    :members:
    :inherited-members:
    :show-inheritance:
+
+
+.. autoclass:: optbinning.binning.outlier.YQuantileDetector
+   :members:
+   :inherited-members:
+   :show-inheritance:

doc/source/release_notes.rst

@@ -1,6 +1,23 @@
 Release Notes
 =============
 
+Version 0.16.1 (2022-11-28)
+---------------------------
+
+New features:
+
+   - Outlier detector``YQuantileDetector`` for continuous target (`Issue 203 <https://github.com/guillermo-navas-palencia/optbinning/issues/203>`_).
+
+Improvements
+
+   - Add support to solver SCS and HIGHS for optimal piecewise binning classes.
+   - Unit testing outlier detector methods.
+
+Bugfixes
+
+   - Pass ``lb`` and ``ub`` as keyword arguments to RoPWR fit method (required since ropwr>=0.4.0).
+
+
 Version 0.16.0 (2022-10-24)
 ---------------------------
 

optbinning/_version.py

@@ -1,3 +1,3 @@
 """Version information."""
 
-__version__ = "0.16.0"
+__version__ = "0.16.1"

optbinning/binning/binning_statistics.py

@@ -617,15 +617,17 @@ def plot(self, metric="woe", add_special=True, add_missing=True,
         add_missing : bool (default=True)
             Whether to add the special values bin.
 
-        style: str, optional (default="bin")
+        style : str, optional (default="bin")
             Plot style. style="bin" shows the standard binning plot. If
             style="actual", show the plot with the actual scale, i.e, actual
             bin widths.
 
-        show_bin_labels: bool (default=False)
+        show_bin_labels : bool (default=False)
             Whether to show the bin label instead of the bin id on the x-axis.
             For long labels (length > 27), labels are truncated.
 
+            .. versionadded:: 0.15.1
+
         savefig : str or None (default=None)
             Path to save the plot figure.
         """
@@ -1203,10 +1205,12 @@ def plot(self, add_special=True, add_missing=True, show_bin_labels=False,
         add_missing : bool (default=True)
             Whether to add the special values bin.
 
-        show_bin_labels: bool (default=False)
+        show_bin_labels : bool (default=False)
             Whether to show the bin label instead of the bin id on the x-axis.
             For long labels (length > 27), labels are truncated.
 
+            .. versionadded:: 0.15.1
+
         savefig : str or None (default=None)
             Path to save the plot figure.
         """
@@ -1660,10 +1664,12 @@ def plot(self, add_special=True, add_missing=True, style="bin",
             style="actual", show the plot with the actual scale, i.e, actual
             bin widths.
 
-        show_bin_labels: bool (default=False)
+        show_bin_labels : bool (default=False)
             Whether to show the bin label instead of the bin id on the x-axis.
             For long labels (length > 27), labels are truncated.
 
+            .. versionadded:: 0.15.1
+
         savefig : str or None (default=None)
             Path to save the plot figure.
         """

optbinning/binning/continuous_binning.py

@@ -118,9 +118,10 @@ def _check_parameters(name, dtype, prebinning_method, max_n_prebins,
         raise ValueError("gamma must be >= 0; got {}.".format(gamma))
 
     if outlier_detector is not None:
-        if outlier_detector not in ("range", "zscore"):
+        if outlier_detector not in ("range", "zscore", "yquantile"):
             raise ValueError('Invalid value for outlier_detector. Allowed '
-                             'string values are "range" and "zscore".')
+                             'string values are "range", "zscore" and '
+                             '"yquantile".')
 
         if outlier_params is not None:
             if not isinstance(outlier_params, dict):
@@ -255,8 +256,9 @@ class ContinuousOptimalBinning(OptimalBinning):
 
     outlier_detector : str or None, optional (default=None)
         The outlier detection method. Supported methods are "range" to use
-        the interquartile range based method or "zcore" to use the modified
-        Z-score method.
+        the interquartile range based method, "zcore" to use the modified
+        Z-score method or "yquantile" to use the y-axis detector over
+        quantiles.
 
     outlier_params : dict or None, optional (default=None)
         Dictionary of parameters to pass to the outlier detection method.

optbinning/binning/outlier.py

@@ -21,18 +21,20 @@ def __init__(self):
         # flag
         self._is_fitted = False
 
-    def fit(self, x):
-        """Fit univariate outlier detector.
+    def fit(self, x, y=None):
+        """Fit outlier detector.
 
         Parameters
         ----------
         x : array-like, shape = (n_samples)
 
+        y : array-like, shape = (n_samples) or None (default=None)
+
         Returns
         -------
         self : OutlierDetector
         """
-        self._fit(x)
+        self._fit(x, y)
 
         return self
 
@@ -88,7 +90,7 @@ def __init__(self, interval_length=0.5, k=1.5, method="ETI"):
         self.k = k
         self.method = method
 
-    def _fit(self, x):
+    def _fit(self, x, y=None):
         if self.method not in ("ETI", "HDI"):
             raise ValueError('Invalid value for method. Allowed string '
                              'values are "ETI" and "HDI".')
@@ -144,7 +146,12 @@ class ModifiedZScoreDetector(BaseEstimator, OutlierDetector):
     def __init__(self, threshold=3.5):
         self.threshold = threshold
 
-    def _fit(self, x):
+    def _fit(self, x, y=None):
+        if (not isinstance(self.threshold, numbers.Number) or
+                self.threshold < 0):
+            raise ValueError("threshold must be a value >= 0; got {}".
+                             format(self.threshold))
+
         x = np.asarray(x)
         median = np.median(x)
         mad = np.median(np.abs(x - median))
@@ -153,3 +160,68 @@ def _fit(self, x):
         self._support = np.abs(m_z_score) > self.threshold
 
         self._is_fitted = True
+
+
+class YQuantileDetector(BaseEstimator, OutlierDetector):
+    """Outlier detector on the y-axis over quantiles.
+
+    Parameters
+    ----------
+    outlier_detector : str or None, optional (default=None)
+        The outlier detection method. Supported methods are "range" to use
+        the interquartile range based method or "zcore" to use the modified
+        Z-score method.
+
+    outlier_params : dict or None, optional (default=None)
+        Dictionary of parameters to pass to the outlier detection method.
+
+    n_bins : int (default=5)
+        The maximum number of bins to consider.
+    """
+    def __init__(self, outlier_detector="zscore",  outlier_params=None,
+                 n_bins=5):
+        self.outlier_detector = outlier_detector
+        self.outlier_params = outlier_params
+        self.n_bins = n_bins
+
+    def _fit(self, x, y):
+        if self.outlier_detector not in ("range", "zscore"):
+            raise ValueError('Invalid value for outlier_detector. Allowed '
+                             'string values are "range" and "zscore".')
+
+        if self.outlier_params is not None:
+            if not isinstance(self.outlier_params, dict):
+                raise TypeError("outlier_params must be a dict or None; "
+                                "got {}.".format(self.outlier_params))
+
+        if not isinstance(self.n_bins, numbers.Integral) or self.n_bins <= 0:
+            raise ValueError("bins must be a positive integer; got {}."
+                             .format(self.n_bins))
+
+        x = np.asarray(x)
+        y = np.asarray(y)
+
+        q = np.linspace(0, 1, self.n_bins + 1)
+        splits = np.unique(np.quantile(x, q))[1:-1]
+        n_bins = len(splits) + 1
+        indices = np.digitize(x, splits, right=False)
+
+        self._support = np.zeros(x.size, dtype=bool)
+        idx_support = np.arange(x.size)
+
+        if self.outlier_detector == "zscore":
+            detector = ModifiedZScoreDetector()
+        elif self.outlier_detector == "range":
+            detector = RangeDetector()
+
+        if self.outlier_params is not None:
+            detector.set_params(**self.outlier_params)
+
+        for i in range(n_bins):
+            mask_x = indices == i
+            detector.fit(y[mask_x])
+            mask_out = detector.get_support()
+            idx_out = idx_support[mask_x][mask_out]
+            self._support[idx_out] = True
+
+        self._is_fitted = True

optbinning/binning/piecewise/base.py

@@ -132,9 +132,10 @@ def _check_parameters(name, estimator, objective, degree, continuous,
                          'values are "all" and "consecutive".')
 
     if outlier_detector is not None:
-        if outlier_detector not in ("range", "zscore"):
+        if outlier_detector not in ("range", "zscore", "yquantile"):
             raise ValueError('Invalid value for outlier_detector. Allowed '
-                             'string values are "range" and "zscore".')
+                             'string values are "range", "zscore" and '
+                             '"yquantile".')
 
         if outlier_params is not None:
             if not isinstance(outlier_params, dict):
@@ -171,9 +172,10 @@ def _check_parameters(name, estimator, objective, degree, continuous,
             raise ValueError("split_digist must be an integer in [0, 8]; "
                              "got {}.".format(split_digits))
 
-    if solver not in ("auto", "ecos", "osqp", "direct"):
+    if solver not in ("auto", "ecos", "osqp", "direct", "scs", "highs"):
         raise ValueError('Invalid value for solver. Allowed string '
-                         'values are "auto", "ecos", "osqp" and "direct".')
+                         'values are "auto", "ecos", "osqp", "direct", '
+                         '"scs" and "highs".')
 
     if not isinstance(h_epsilon, numbers.Number) or h_epsilon < 1.0:
         raise ValueError("h_epsilon must a number >= 1.0; got {}."
@@ -453,7 +455,7 @@ def _fit_binning(self, x, y, prediction, lb, ub):
             self.solver, self.h_epsilon, self.quantile, self.regularization,
             self.reg_l1, self.reg_l1, self.verbose)
 
-        optimizer.fit(x_subsamples, pred_subsamples, splits, lb, ub)
+        optimizer.fit(x_subsamples, pred_subsamples, splits, lb=lb, ub=ub)
 
         self._c = optimizer.coef_
 

optbinning/binning/piecewise/binning.py

@@ -114,8 +114,9 @@ class OptimalPWBinning(BasePWBinning):
 
     outlier_detector : str or None, optional (default=None)
         The outlier detection method. Supported methods are "range" to use
-        the interquartile range based method or "zcore" to use the modified
-        Z-score method.
+        the interquartile range based method, "zcore" to use the modified
+        Z-score method or "yquantile" to use the y-axis detector over
+        quantiles.
 
     outlier_params : dict or None, optional (default=None)
         Dictionary of parameters to pass to the outlier detection method.
@@ -142,7 +143,9 @@ class OptimalPWBinning(BasePWBinning):
         <https://github.com/embotech/ecos>`_, `"osqp"
         <https://github.com/oxfordcontrol/osqp>`_, "direct", to choose the
         direct solver, and "auto", to choose the most appropriate solver for
-        the problem.
+        the problem. Version 0.16.1 added support to solvers
+        `"scs" <https://github.com/cvxgrp/scs>`_ and `"highs"
+        <https://github.com/ERGO-Code/HiGHS>`_.
 
     h_epsilon: float (default=1.35)
         The parameter h_epsilon used when ``objective="huber"``, controls the

optbinning/binning/piecewise/continuous_binning.py

@@ -104,8 +104,9 @@ class ContinuousOptimalPWBinning(BasePWBinning):
 
     outlier_detector : str or None, optional (default=None)
         The outlier detection method. Supported methods are "range" to use
-        the interquartile range based method or "zcore" to use the modified
-        Z-score method.
+        the interquartile range based method, "zcore" to use the modified
+        Z-score method or "yquantile" to use the y-axis detector over
+        quantiles.
 
     outlier_params : dict or None, optional (default=None)
         Dictionary of parameters to pass to the outlier detection method.
@@ -132,7 +133,9 @@ class ContinuousOptimalPWBinning(BasePWBinning):
         <https://github.com/embotech/ecos>`_, `"osqp"
         <https://github.com/oxfordcontrol/osqp>`_, "direct", to choose the
         direct solver, and "auto", to choose the most appropriate solver for
-        the problem.
+        the problem. Version 0.16.1 added support to solvers
+        `"scs" <https://github.com/cvxgrp/scs>`_ and `"highs"
+        <https://github.com/ERGO-Code/HiGHS>`_.
 
     h_epsilon: float (default=1.35)
         The parameter h_epsilon used when ``objective="huber"``, controls the
@@ -340,15 +343,6 @@ def _fit(self, x, y, lb, ub, check_input):
 
         time_postprocessing = time.perf_counter()
 
-        # Compute n_records and sum for special and missing
-        # self._n_records_special = len(y_special)
-        # self._sum_special = np.sum(y_special)
-        # self._n_zeros_special = np.count_nonzero(y_special == 0)
-        # if len(y_special):
-        #     self._std_special = np.std(y_special)
-        #     self._min_target_special = np.min(y_special)
-        #     self._max_target_special = np.max(y_special)
-
         [self._n_records_special, self._sum_special, self._n_zeros_special,
          self._std_special, self._min_target_special,
          self._max_target_special] = target_info_special_continuous(

optbinning/binning/preprocessing.py

@@ -18,6 +18,7 @@
 
 from .outlier import ModifiedZScoreDetector
 from .outlier import RangeDetector
+from .outlier import YQuantileDetector
 
 
 def categorical_transform(x, y):
@@ -78,8 +79,9 @@ def split_data(dtype, x, y, special_codes=None, cat_cutoff=None,
 
     outlier_detector : str or None (default=None)
         The outlier detection method. Supported methods are "range" to use
-        the interquartile range based method or "zcore" to use the modified
-        Z-score method.
+        the interquartile range based method, "zcore" to use the modified
+        Z-score method or "yquantile" to use the y-axis detector over
+        quantiles.
 
     outlier_params : dict or None (default=None)
         Dictionary of parameters to pass to the outlier detection method.
@@ -139,9 +141,10 @@ def split_data(dtype, x, y, special_codes=None, cat_cutoff=None,
         Others data sample weight.
     """
     if outlier_detector is not None:
-        if outlier_detector not in ("range", "zscore"):
+        if outlier_detector not in ("range", "zscore", "yquantile"):
             raise ValueError('Invalid value for outlier_detector. Allowed '
-                             'string values are "range" and "zscore".')
+                             'string values are "range", "zscore" and '
+                             '"yquantile".')
 
         if outlier_params is not None:
             if not isinstance(outlier_params, dict):
@@ -229,11 +232,17 @@ def split_data(dtype, x, y, special_codes=None, cat_cutoff=None,
                 detector = RangeDetector()
             elif outlier_detector == "zscore":
                 detector = ModifiedZScoreDetector()
+            elif outlier_detector == "yquantile":
+                detector = YQuantileDetector()
 
             if outlier_params is not None:
                 detector.set_params(**outlier_params)
 
-            mask_outlier = detector.fit(x_clean).get_support()
+            if outlier_detector in ("range", "zscore"):
+                mask_outlier = detector.fit(x_clean).get_support()
+            else:
+                mask_outlier = detector.fit(x_clean, y_clean).get_support()
+
             x_clean = x_clean[~mask_outlier]
             y_clean = y_clean[~mask_outlier]
             sw_clean = sw_clean[~mask_outlier]