Update sklearn version

hiclass/BinaryPolicy.py

@@ -163,7 +163,7 @@ def get_binary_examples(self, node) -> tuple:
         elif isinstance(self.X, csr_matrix) or isinstance(self.X, csr_array):
             X = vstack([positive_x, negative_x])
             sample_weights = (
-                vstack([positive_weights, negative_weights])
+                np.concatenate([positive_weights, negative_weights])
                 if self.sample_weight is not None
                 else None
             )

hiclass/HierarchicalClassifier.py

@@ -20,6 +20,10 @@
     MultiplyCombiner,
 )
 
+from hiclass.probability_combiner import (
+    init_strings as probability_combiner_init_strings,
+)
+
 try:
     import ray
 except ImportError:
@@ -79,6 +83,7 @@ def __init__(
         bert: bool = False,
         classifier_abbreviation: str = "",
         calibration_method: str = None,
+        probability_combiner: str = "multiply",
         tmp_dir: str = None,
     ):
         """
@@ -107,6 +112,13 @@ def __init__(
             The abbreviation of the local hierarchical classifier to be displayed during logging.
         calibration_method : {"ivap", "cvap", "platt", "isotonic", "beta"}, str, default=None
             If set, use the desired method to calibrate probabilities returned by predict_proba().
+        probability_combiner: {"geometric", "arithmetic", "multiply", None}, str, default="multiply"
+            Specify the rule for combining probabilities over multiple levels:
+
+            - `geometric`: Each levels probabilities are calculated by taking the geometric mean of itself and its predecessors;
+            - `arithmetic`: Each levels probabilities are calculated by taking the arithmetic mean of itself and its predecessors;
+            - `multiply`: Each levels probabilities are calculated by multiplying itself with its predecessors.
+            - `None`: No aggregation.
         tmp_dir : str, default=None
             Temporary directory to persist local classifiers that are trained. If the job needs to be restarted,
             it will skip the pre-trained local classifier found in the temporary directory.
@@ -119,6 +131,7 @@ def __init__(
         self.bert = bert
         self.classifier_abbreviation = classifier_abbreviation
         self.calibration_method = calibration_method
+        self.probability_combiner = probability_combiner
         self.tmp_dir = tmp_dir
 
     def fit(self, X, y, sample_weight=None):
@@ -152,6 +165,15 @@ def fit(self, X, y, sample_weight=None):
             self._clean_up()
 
     def _pre_fit(self, X, y, sample_weight):
+        # check params
+        if (
+            self.probability_combiner
+            and self.probability_combiner not in probability_combiner_init_strings
+        ):
+            raise ValueError(
+                f"probability_combiner must be one of {', '.join(probability_combiner_init_strings)} or None."
+            )
+
         # Check that X and y have correct shape
         # and convert them to np.ndarray if need be
 

hiclass/LocalClassifierPerLevel.py

@@ -12,15 +12,12 @@
 import numpy as np
 from joblib import Parallel, delayed
 from sklearn.base import BaseEstimator
+from sklearn.utils._tags import ClassifierTags
 from sklearn.utils.validation import check_array, check_is_fitted
 
+from hiclass._calibration.Calibrator import _Calibrator
 from hiclass.ConstantClassifier import ConstantClassifier
 from hiclass.HierarchicalClassifier import HierarchicalClassifier
-from hiclass._calibration.Calibrator import _Calibrator
-
-from hiclass.probability_combiner import (
-    init_strings as probability_combiner_init_strings,
-)
 
 try:
     import ray
@@ -113,13 +110,12 @@ def __init__(
         self.return_all_probabilities = return_all_probabilities
         self.probability_combiner = probability_combiner
 
-        if (
-            self.probability_combiner
-            and self.probability_combiner not in probability_combiner_init_strings
-        ):
-            raise ValueError(
-                f"probability_combiner must be one of {', '.join(probability_combiner_init_strings)} or None."
-            )
+    def __sklearn_tags__(self):
+        """Configure annotations of estimator to allow inspection of capabilities, such as sparse matrix support."""
+        tags = super().__sklearn_tags__()
+        tags.input_tags.sparse = True
+        tags.classifier_tags = ClassifierTags()
+        return tags
 
     def fit(self, X, y, sample_weight=None):
         """

hiclass/LocalClassifierPerNode.py

@@ -12,18 +12,14 @@
 import networkx as nx
 import numpy as np
 from sklearn.base import BaseEstimator
+from sklearn.utils._tags import ClassifierTags
 from sklearn.utils.validation import check_array, check_is_fitted
 
 from hiclass import BinaryPolicy
-from hiclass.ConstantClassifier import ConstantClassifier
-from hiclass.HierarchicalClassifier import HierarchicalClassifier
 from hiclass._calibration.Calibrator import _Calibrator
-
-from hiclass.probability_combiner import (
-    init_strings as probability_combiner_init_strings,
-)
-
 from hiclass._hiclass_utils import _normalize_probabilities
+from hiclass.ConstantClassifier import ConstantClassifier
+from hiclass.HierarchicalClassifier import HierarchicalClassifier
 
 
 class LocalClassifierPerNode(BaseEstimator, HierarchicalClassifier):
@@ -122,13 +118,12 @@ def __init__(
         self.return_all_probabilities = return_all_probabilities
         self.probability_combiner = probability_combiner
 
-        if (
-            self.probability_combiner
-            and self.probability_combiner not in probability_combiner_init_strings
-        ):
-            raise ValueError(
-                f"probability_combiner must be one of {', '.join(probability_combiner_init_strings)} or None."
-            )
+    def __sklearn_tags__(self):
+        """Configure annotations of estimator to allow inspection of capabilities, such as sparse matrix support."""
+        tags = super().__sklearn_tags__()
+        tags.input_tags.sparse = True
+        tags.classifier_tags = ClassifierTags()
+        return tags
 
     def fit(self, X, y, sample_weight=None):
         """

hiclass/LocalClassifierPerParentNode.py

@@ -12,17 +12,13 @@
 import networkx as nx
 import numpy as np
 from sklearn.base import BaseEstimator
+from sklearn.utils._tags import ClassifierTags
 from sklearn.utils.validation import check_array, check_is_fitted
 
-from hiclass.ConstantClassifier import ConstantClassifier
-from hiclass.HierarchicalClassifier import HierarchicalClassifier
 from hiclass._calibration.Calibrator import _Calibrator
-
-from hiclass.probability_combiner import (
-    init_strings as probability_combiner_init_strings,
-)
-
 from hiclass._hiclass_utils import _normalize_probabilities
+from hiclass.ConstantClassifier import ConstantClassifier
+from hiclass.HierarchicalClassifier import HierarchicalClassifier
 
 
 class LocalClassifierPerParentNode(BaseEstimator, HierarchicalClassifier):
@@ -108,13 +104,12 @@ def __init__(
         self.return_all_probabilities = return_all_probabilities
         self.probability_combiner = probability_combiner
 
-        if (
-            self.probability_combiner
-            and self.probability_combiner not in probability_combiner_init_strings
-        ):
-            raise ValueError(
-                f"probability_combiner must be one of {', '.join(probability_combiner_init_strings)} or None."
-            )
+    def __sklearn_tags__(self):
+        """Configure annotations of estimator to allow inspection of capabilities, such as sparse matrix support."""
+        tags = super().__sklearn_tags__()
+        tags.input_tags.sparse = True
+        tags.classifier_tags = ClassifierTags()
+        return tags
 
     def fit(self, X, y, sample_weight=None):
         """

hiclass/MultiLabelHierarchicalClassifier.py

@@ -11,14 +11,6 @@
 from sklearn.linear_model import LogisticRegression
 from sklearn.utils.validation import _check_sample_weight
 
-import functools
-import sklearn.utils.validation
-
-# TODO: Move to MultiLabelHierarchicalClassifier (Parent Class)
-sklearn.utils.validation.check_array = functools.partial(
-    sklearn.utils.validation.check_array, allow_nd=True
-)
-
 try:
     import ray
 except ImportError:

hiclass/MultiLabelLocalClassifierPerNode.py

@@ -4,11 +4,16 @@
 Numeric and string output labels are both handled.
 """
 
+# monkeypatching check_array to accept 3 dimensional arrays
+import functools
 from copy import deepcopy
 
-import functools
 import networkx as nx
 import numpy as np
+import sklearn.utils.validation
+from sklearn.base import BaseEstimator
+from sklearn.utils._tags import ClassifierTags, TargetTags
+from sklearn.utils.validation import check_is_fitted
 
 from hiclass import BinaryPolicy
 from hiclass.ConstantClassifier import ConstantClassifier
@@ -17,13 +22,6 @@
     make_leveled,
 )
 
-from sklearn.base import BaseEstimator
-from sklearn.utils.validation import check_is_fitted
-
-# monkeypatching check_array to accept 3 dimensional arrays
-import sklearn.utils.validation
-
-# TODO: Move to MultiLabelHierarchicalClassifier (Parent Class)
 sklearn.utils.validation.check_array = functools.partial(
     sklearn.utils.validation.check_array, allow_nd=True
 )
@@ -108,6 +106,16 @@ def __init__(
         self.binary_policy = binary_policy
         self.tolerance = tolerance
 
+    def __sklearn_tags__(self):
+        """Configure annotations of estimator to allow inspection of capabilities, such as sparse matrix support."""
+        tags = super().__sklearn_tags__()
+        tags.input_tags.sparse = True
+        tags.classifier_tags = ClassifierTags()
+        tags.target_tags = TargetTags(required=True)
+        tags.target_tags.multi_output = True
+        tags.target_tags.single_output = False
+        return tags
+
     def fit(self, X, y, sample_weight=None):
         """
         Fit a local classifier per node.
@@ -175,9 +183,7 @@ def predict(self, X, tolerance: float = None) -> np.ndarray:
 
         # Input validation
         if not self.bert:
-            X = sklearn.utils.validation.check_array(
-                X, accept_sparse="csr"
-            )  # TODO: Decide allow_nd True or False
+            X = sklearn.utils.validation.check_array(X, accept_sparse="csr")
         else:
             X = np.array(X)
 

hiclass/MultiLabelLocalClassifierPerParentNode.py

@@ -4,21 +4,29 @@
 Numeric and string output labels are both handled.
 """
 
-from copy import deepcopy
+# monkeypatching check_array to accept 3 dimensional arrays
+import functools
 from collections import defaultdict
+from copy import deepcopy
 
 import networkx as nx
 import numpy as np
+import sklearn.utils.validation
 from scipy.sparse import csr_matrix, vstack
 from sklearn.base import BaseEstimator
-from sklearn.utils.validation import check_array, check_is_fitted
+from sklearn.utils._tags import ClassifierTags, TargetTags
+from sklearn.utils.validation import check_is_fitted
 
 from hiclass.ConstantClassifier import ConstantClassifier
 from hiclass.MultiLabelHierarchicalClassifier import (
     MultiLabelHierarchicalClassifier,
     make_leveled,
 )
 
+sklearn.utils.validation.check_array = functools.partial(
+    sklearn.utils.validation.check_array, allow_nd=True
+)
+
 
 class MultiLabelLocalClassifierPerParentNode(
     BaseEstimator, MultiLabelHierarchicalClassifier
@@ -88,6 +96,14 @@ def __init__(
             bert=bert,
         )
 
+    def __sklearn_tags__(self):
+        """Configure annotations of estimator to allow inspection of capabilities, such as sparse matrix support."""
+        tags = super().__sklearn_tags__()
+        tags.input_tags.sparse = True
+        tags.classifier_tags = ClassifierTags()
+        tags.target_tags = TargetTags(required=True)
+        return tags
+
     def fit(self, X, y, sample_weight=None):
         """
         Fit a local classifier per parent node.
@@ -152,7 +168,7 @@ def predict(self, X, tolerance: float = None):
 
         # Input validation
         if not self.bert:
-            X = check_array(X, accept_sparse="csr")
+            X = sklearn.utils.validation.check_array(X, accept_sparse="csr")
         else:
             X = np.array(X)
 

setup.py

@@ -27,13 +27,13 @@
 KEYWORDS = ["hierarchical classification"]
 DACS_SOFTWARE = "https://gitlab.com/dacs-hpi"
 # What packages are required for this module to be executed?
-REQUIRED = ["networkx", "numpy", "scikit-learn<1.5", "scipy<1.13"]
+REQUIRED = ["networkx", "numpy", "scikit-learn", "scipy"]
 
 # What packages are optional?
 # 'fancy feature': ['django'],}
 EXTRAS = {
     "ray": ["ray>=1.11.0"],
-    "xai": ["shap==0.44.1", "xarray==2023.1.0"],
+    "xai": ["shap", "xarray"],
     "dev": [
         "flake8",
         "pytest",
@@ -43,8 +43,8 @@
         "black==24.2.0",
         "pre-commit==2.20.0",
         "ray",
-        "shap==0.44.1",
-        "xarray==2023.1.0",
+        "shap",
+        "xarray",
         "bert-sklearn",
     ],
 }