Merge pull request #354 from KhiopsML/support-float-and-bool-targets-in-khiops-classifiers

Support float and bool targets in khiops classifiers

Author: nairbenrekia

CHANGELOG.md

@@ -6,6 +6,11 @@
   - Example: 10.2.1.4 is the 5th version that supports khiops 10.2.1.
 - Internals: Changes in *Internals* sections are unlikely to be of interest for data scientists.
 
+## Unreleased
+
+### Added
+- (`sklearn`) Support for boolean and float targets in `KhiopsClassifier`.
+
 ## 10.3.0.0 - 2025-02-10
 
 ### Fixed

khiops/sklearn/dataset.py

@@ -14,6 +14,7 @@
 
 import numpy as np
 import pandas as pd
+import sklearn
 from scipy import sparse as sp
 from sklearn.utils import check_array
 from sklearn.utils.validation import column_or_1d
@@ -430,6 +431,19 @@ def write_internal_data_table(dataframe, file_path_or_stream):
     )
 
 
+def _column_or_1d_with_dtype(y, dtype=None):
+    # 'dtype' has been introduced on `column_or_1d' since Scikit-learn 1.2;
+    if sklearn.__version__ < "1.2":
+        if pd.api.types.is_string_dtype(dtype) and y.isin(["True", "False"]).all():
+            warnings.warn(
+                "'y' stores strings restricted to 'True'/'False' values: "
+                "The predict method may return a bool vector."
+            )
+        return column_or_1d(y, warn=True)
+    else:
+        return column_or_1d(y, warn=True, dtype=dtype)
+
+
 class Dataset:
     """A representation of a dataset
 
@@ -738,8 +752,22 @@ def _init_target_column(self, y):
         if isinstance(y, str):
             y_checked = y
         else:
-            y_checked = column_or_1d(y, warn=True)
-
+            if hasattr(y, "dtype"):
+                if isinstance(y.dtype, pd.CategoricalDtype):
+                    y_checked = _column_or_1d_with_dtype(
+                        y, dtype=y.dtype.categories.dtype
+                    )
+                else:
+                    y_checked = _column_or_1d_with_dtype(y, dtype=y.dtype)
+            elif hasattr(y, "dtypes"):
+                if isinstance(y.dtypes[0], pd.CategoricalDtype):
+                    y_checked = _column_or_1d_with_dtype(
+                        y, dtype=y.dtypes[0].categories.dtype
+                    )
+                else:
+                    y_checked = _column_or_1d_with_dtype(y)
+            else:
+                y_checked = _column_or_1d_with_dtype(y)
         # Check the target type coherence with those of X's tables
         if isinstance(
             self.main_table, (PandasTable, SparseTable, NumpyTable)

khiops/sklearn/estimators.py

@@ -154,6 +154,7 @@ def _check_categorical_target_type(ds):
         or pd.api.types.is_string_dtype(ds.target_column.dtype)
         or pd.api.types.is_integer_dtype(ds.target_column.dtype)
         or pd.api.types.is_float_dtype(ds.target_column.dtype)
+        or pd.api.types.is_bool_dtype(ds.target_column.dtype)
     ):
         raise ValueError(
             f"'y' has invalid type '{ds.target_column_type}'. "
@@ -2123,6 +2124,24 @@ def _is_real_target_dtype_integer(self):
             )
         )
 
+    def _is_real_target_dtype_float(self):
+        return self._original_target_dtype is not None and (
+            pd.api.types.is_float_dtype(self._original_target_dtype)
+            or (
+                isinstance(self._original_target_dtype, pd.CategoricalDtype)
+                and pd.api.types.is_float_dtype(self._original_target_dtype.categories)
+            )
+        )
+
+    def _is_real_target_dtype_bool(self):
+        return self._original_target_dtype is not None and (
+            pd.api.types.is_bool_dtype(self._original_target_dtype)
+            or (
+                isinstance(self._original_target_dtype, pd.CategoricalDtype)
+                and pd.api.types.is_bool_dtype(self._original_target_dtype.categories)
+            )
+        )
+
     def _sorted_prob_variable_names(self):
         """Returns the model probability variable names in the order of self.classes_"""
         self._assert_is_fitted()
@@ -2227,8 +2246,13 @@ def _fit_training_post_process(self, ds):
             for key in variable.meta_data.keys:
                 if key.startswith("TargetProb"):
                     self.classes_.append(variable.meta_data.get_value(key))
-        if ds.is_in_memory and self._is_real_target_dtype_integer():
-            self.classes_ = [int(class_value) for class_value in self.classes_]
+        if ds.is_in_memory:
+            if self._is_real_target_dtype_integer():
+                self.classes_ = [int(class_value) for class_value in self.classes_]
+            elif self._is_real_target_dtype_float():
+                self.classes_ = [float(class_value) for class_value in self.classes_]
+            elif self._is_real_target_dtype_bool():
+                self.classes_ = [class_value == "True" for class_value in self.classes_]
             self.classes_.sort()
         self.classes_ = column_or_1d(self.classes_)
 
@@ -2283,9 +2307,10 @@ def predict(self, X):
         -------
         `ndarray <numpy.ndarray>`
             An array containing the encoded columns. A first column containing key
-            column ids is added in multi-table mode. The `numpy.dtype` of the array is
-            integer if the classifier was learned with an integer ``y``. Otherwise it
-            will be ``str``.
+            column ids is added in multi-table mode. The `numpy.dtype` of the array
+            matches the type of ``y`` used during training. It will be integer, float,
+            or boolean if the classifier was trained with a ``y`` of the corresponding
+            type. Otherwise it will be ``str``.
 
             The key columns are added for multi-table tasks.
         """

tests/test_sklearn_output_types.py

@@ -71,12 +71,12 @@ def test_classifier_output_types(self):
         khc = KhiopsClassifier(n_trees=0)
         khc.fit(X, y)
         y_pred = khc.predict(X)
+        khc.fit(X_mt, y)
+        y_mt_pred = khc.predict(X_mt)
+
         y_bin = y.replace({0: 0, 1: 0, 2: 1})
         khc.fit(X, y_bin)
         y_bin_pred = khc.predict(X)
-        khc.fit(X_mt, y)
-        khc.export_report_file("report.khj")
-        y_mt_pred = khc.predict(X_mt)
         khc.fit(X_mt, y_bin)
         y_mt_bin_pred = khc.predict(X_mt)
 
@@ -85,6 +85,8 @@ def test_classifier_output_types(self):
             "ys": {
                 "int": y,
                 "int binary": y_bin,
+                "float": y.astype(float),
+                "bool": y.replace({0: True, 1: True, 2: False}),
                 "string": self._replace(y, {0: "se", 1: "vi", 2: "ve"}),
                 "string binary": self._replace(y_bin, {0: "vi_or_se", 1: "ve"}),
                 "int as string": self._replace(y, {0: "8", 1: "9", 2: "10"}),
@@ -93,30 +95,42 @@ def test_classifier_output_types(self):
                 "cat string": pd.Series(
                     self._replace(y, {0: "se", 1: "vi", 2: "ve"})
                 ).astype("category"),
+                "cat float": y.astype(float).astype("category"),
+                "cat bool": y.replace({0: True, 1: True, 2: False}).astype("category"),
             },
             "y_type_check": {
                 "int": pd.api.types.is_integer_dtype,
                 "int binary": pd.api.types.is_integer_dtype,
+                "float": pd.api.types.is_float_dtype,
+                "bool": pd.api.types.is_bool_dtype,
                 "string": pd.api.types.is_string_dtype,
                 "string binary": pd.api.types.is_string_dtype,
                 "int as string": pd.api.types.is_string_dtype,
                 "int as string binary": pd.api.types.is_string_dtype,
                 "cat int": pd.api.types.is_integer_dtype,
                 "cat string": pd.api.types.is_string_dtype,
+                "cat float": pd.api.types.is_float_dtype,
+                "cat bool": pd.api.types.is_bool_dtype,
             },
             "expected_classes": {
                 "int": column_or_1d([0, 1, 2]),
                 "int binary": column_or_1d([0, 1]),
+                "float": column_or_1d([0.0, 1.0, 2.0]),
+                "bool": column_or_1d([False, True]),
                 "string": column_or_1d(["se", "ve", "vi"]),
                 "string binary": column_or_1d(["ve", "vi_or_se"]),
                 "int as string": column_or_1d(["10", "8", "9"]),
                 "int as string binary": column_or_1d(["10", "89"]),
                 "cat int": column_or_1d([0, 1, 2]),
                 "cat string": column_or_1d(["se", "ve", "vi"]),
+                "cat float": column_or_1d([0.0, 1.0, 2.0]),
+                "cat bool": column_or_1d([False, True]),
             },
             "expected_y_preds": {
                 "mono": {
                     "int": y_pred,
+                    "float": y_pred.astype(float),
+                    "bool": self._replace(y_bin_pred, {0: True, 1: False}),
                     "int binary": y_bin_pred,
                     "string": self._replace(y_pred, {0: "se", 1: "vi", 2: "ve"}),
                     "string binary": self._replace(
@@ -128,9 +142,15 @@ def test_classifier_output_types(self):
                     ),
                     "cat int": y_pred,
                     "cat string": self._replace(y_pred, {0: "se", 1: "vi", 2: "ve"}),
+                    "cat float": self._replace(
+                        y_pred, {target: float(target) for target in (0, 1, 2)}
+                    ),
+                    "cat bool": self._replace(y_bin_pred, {0: True, 1: False}),
                 },
                 "multi": {
                     "int": y_mt_pred,
+                    "float": y_mt_pred.astype(float),
+                    "bool": self._replace(y_mt_bin_pred, {0: True, 1: False}),
                     "int binary": y_mt_bin_pred,
                     "string": self._replace(y_mt_pred, {0: "se", 1: "vi", 2: "ve"}),
                     "string binary": self._replace(
@@ -144,6 +164,10 @@ def test_classifier_output_types(self):
                     ),
                     "cat int": y_mt_pred,
                     "cat string": self._replace(y_mt_pred, {0: "se", 1: "vi", 2: "ve"}),
+                    "cat float": self._replace(
+                        y_mt_pred, {target: float(target) for target in (0, 1, 2)}
+                    ),
+                    "cat bool": self._replace(y_mt_bin_pred, {0: True, 1: False}),
                 },
             },
             "Xs": {