added parameters for specifying the number of labels in nominal attri…

…butes/class to avoid "cannot handle unary class attribute" if particular split only contains one label

CHANGES.rst

@@ -4,7 +4,10 @@ Changelog
 0.0.6 (????-??-??)
 ------------------
 
-- ...
+- `WekaEstimator` (module `sklweka.classifiers`) and `WekaCluster` (module `sklweka.clusters`)
+  now allow specifying how many labels a particular nominal attribute or class attribute has
+  (to avoid error message like `Cannot handle unary class attribute!` if there is only one
+  label present in a particular split)
 
 
 0.0.5 (2022-04-01)

src/sklweka/classifiers.py

@@ -16,7 +16,8 @@ class WekaEstimator(BaseEstimator, OptionHandler, RegressorMixin, ClassifierMixi
     """
 
     def __init__(self, jobject=None, classifier=None, classname=None, options=None,
-                 nominal_input_vars=None, nominal_output_var=None):
+                 nominal_input_vars=None, nominal_output_var=None,
+                 num_nominal_input_labels=None, num_nominal_output_labels=None):
         """
         Initializes the estimator. Can be either instantiated via the following priority of parameters:
         1. JB_Object representing a Java Classifier object
@@ -35,6 +36,10 @@ def __init__(self, jobject=None, classifier=None, classname=None, options=None,
         :type nominal_input_vars: list or str
         :param nominal_output_var: whether to convert the output variable to a nominal one
         :type nominal_output_var: bool
+        :param num_nominal_input_labels: the dictionary with the number of labels for the nominal input variables (key is 0-based attribute index)
+        :type num_nominal_input_labels: dict
+        :param num_nominal_output_labels: the number of labels for the output variable
+        :type num_nominal_output_labels: int
         """
         if jobject is not None:
             _jobject = jobject
@@ -60,6 +65,8 @@ def __init__(self, jobject=None, classifier=None, classname=None, options=None,
         self._options = options
         self._nominal_input_vars = nominal_input_vars
         self._nominal_output_var = nominal_output_var
+        self._num_nominal_input_labels = num_nominal_input_labels
+        self._num_nominal_output_labels = num_nominal_output_labels
 
     @property
     def classifier(self):
@@ -97,7 +104,9 @@ def fit(self, data, targets):
             data = to_nominal_attributes(data, self._nominal_input_vars)
         if self._nominal_output_var is not None:
             targets = to_nominal_labels(targets)
-        d = to_instances(data, targets)
+        d = to_instances(data, targets,
+                         num_nominal_labels=self._num_nominal_input_labels,
+                         num_class_labels=self._num_nominal_output_labels)
         self._classifier.build_classifier(d)
         self.header_ = d.template_instances(d, 0)
         if d.class_attribute.is_nominal:
@@ -162,6 +171,10 @@ def get_params(self, deep=True):
             result["nominal_input_vars"] = self._nominal_input_vars
         if self._nominal_output_var is not None:
             result["nominal_output_var"] = self._nominal_output_var
+        if self._num_nominal_input_labels is not None:
+            result["num_nominal_input_labels"] = self._num_nominal_input_labels
+        if self._num_nominal_output_labels is not None:
+            result["num_nominal_output_labels"] = self._num_nominal_output_labels
         return result
 
     def set_params(self, **params):
@@ -186,6 +199,12 @@ def set_params(self, **params):
         self._nominal_output_var = None
         if "nominal_output_var" in params:
             self._nominal_output_var = params["nominal_output_var"]
+        self._num_nominal_input_labels = None
+        if "num_nominal_input_labels" in params:
+            self._num_nominal_input_labels = params["num_nominal_input_labels"]
+        self._num_nominal_output_labels = None
+        if "num_nominal_output_labels" in params:
+            self._num_nominal_output_labels = params["num_nominal_output_labels"]
 
     def __str__(self):
         """

src/sklweka/clusters.py

@@ -14,7 +14,8 @@ class WekaCluster(BaseEstimator, OptionHandler, ClusterMixin):
     Wraps a Weka cluster within the scikit-learn framework.
     """
 
-    def __init__(self, jobject=None, cluster=None, classname=None, options=None, nominal_input_vars=None):
+    def __init__(self, jobject=None, cluster=None, classname=None, options=None, nominal_input_vars=None,
+                 num_nominal_input_labels=None):
         """
         Initializes the estimator. Can be either instantiated via the following priority of parameters:
         1. JB_Object representing a Java Clusterer object
@@ -29,8 +30,8 @@ def __init__(self, jobject=None, cluster=None, classname=None, options=None, nom
         :type classname: str
         :param options: the command-line options of the Weka cluster to instantiate
         :type options: list
-        :param nominal_input_vars: the list of 0-based indices of attributes to convert to nominal or range string with 1-based indices
-        :type nominal_input_vars: list or str
+        :param num_nominal_input_labels: the dictionary with the number of labels for the nominal input variables (key is 0-based attribute index)
+        :type num_nominal_input_labels: dict
         """
         if jobject is not None:
             _jobject = jobject
@@ -54,6 +55,7 @@ def __init__(self, jobject=None, cluster=None, classname=None, options=None, nom
         self._classname = classname
         self._options = options
         self._nominal_input_vars = nominal_input_vars
+        self._num_nominal_input_labels = num_nominal_input_labels
 
     @property
     def cluster(self):
@@ -88,7 +90,7 @@ def fit(self, data, targets=None):
         """
         if self._nominal_input_vars is not None:
             data = to_nominal_attributes(data, self._nominal_input_vars)
-        d = to_instances(data)
+        d = to_instances(data, num_nominal_labels=self._num_nominal_input_labels)
         self._cluster.build_clusterer(d)
         self.header_ = d.template_instances(d, 0)
         return self
@@ -141,8 +143,10 @@ def get_params(self, deep=True):
         result["options"] = self._options
         if self._nominal_input_vars is not None:
             result["nominal_input_vars"] = self._nominal_input_vars
-        if self._nominal_output_var is not None:
-            result["nominal_output_var"] = self._nominal_output_var
+        if self._num_nominal_input_labels is not None:
+            result["num_nominal_input_labels"] = self._num_nominal_input_labels
+        if self._num_nominal_input_labels is not None:
+            result["num_nominal_input_labels"] = self._num_nominal_input_labels
         return result
 
     def set_params(self, **params):
@@ -164,9 +168,9 @@ def set_params(self, **params):
         self._nominal_input_vars = None
         if "nominal_input_vars" in params:
             self._nominal_input_vars = params["nominal_input_vars"]
-        self._nominal_output_var = None
-        if "nominal_output_var" in params:
-            self._nominal_output_var = params["nominal_output_var"]
+        self._num_nominal_input_labels = None
+        if "num_nominal_input_labels" in params:
+            self._num_nominal_input_labels = params["num_nominal_input_labels"]
 
     def __str__(self):
         """

src/sklweka/dataset.py

@@ -265,7 +265,8 @@ def determine_attribute_type(y):
     return result
 
 
-def to_instances(X, y=None, att_names=None, att_types=None, class_name=None, class_type=None, relation_name=None):
+def to_instances(X, y=None, att_names=None, att_types=None, class_name=None, class_type=None, relation_name=None,
+                 num_nominal_labels=None, num_class_labels=None):
     """
     Turns the 2D matrix and the optional 1D class vector into an Instances object.
 
@@ -282,6 +283,10 @@ def to_instances(X, y=None, att_names=None, att_types=None, class_name=None, cla
     :type class_type: str
     :param relation_name: the name for the dataset
     :type relation_name: str
+    :param num_nominal_labels: the dictionary with the number of labels (key is 0-based attribute index)
+    :type num_nominal_labels: dict
+    :param num_class_labels: the number of labels in the class attribute
+    :type num_class_labels: int
     :return: the generated Instances object
     :rtype: Instances
     """
@@ -317,12 +322,17 @@ def to_instances(X, y=None, att_names=None, att_types=None, class_name=None, cla
         if att_type == "N":
             atts.append(Attribute.create_numeric(att_name))
         elif att_type == "C":
-            labels = set()
-            for n in range(len(X)):
-                r = X[n]
-                v = str(r[i])
-                labels.add(v)
-            values = sorted(labels)
+            if (num_nominal_labels is not None) and (i in num_nominal_labels):
+                values = []
+                for l in range(num_nominal_labels[i]):
+                    values.append("_%d" % l)
+            else:
+                labels = set()
+                for n in range(len(X)):
+                    r = X[n]
+                    v = str(r[i])
+                    labels.add(v)
+                values = sorted(labels)
             atts.append(Attribute.create_nominal(att_name, values))
         else:
             raise Exception("Unsupported attribute type for column %d: %s" % ((i+1), att_type))
@@ -331,7 +341,12 @@ def to_instances(X, y=None, att_names=None, att_types=None, class_name=None, cla
         if class_type == "N":
             atts.append(Attribute.create_numeric(class_name))
         elif class_type == "C":
-            values = sorted(set([str(x) for x in y]))
+            if num_class_labels is not None:
+                values = []
+                for l in range(num_class_labels):
+                    values.append("_%d" % l)
+            else:
+                values = sorted(set([str(x) for x in y]))
             atts.append(Attribute.create_nominal(class_name, values))
 
     result = Instances.create_instances(relation_name, atts, len(X))

src/sklweka/preprocessing.py

@@ -14,7 +14,8 @@ class WekaTransformer(BaseEstimator, OptionHandler, TransformerMixin):
     Wraps a Weka filter within the scikit-learn framework.
     """
 
-    def __init__(self, jobject=None, filter=None, classname=None, options=None):
+    def __init__(self, jobject=None, filter=None, classname=None, options=None,
+                 num_nominal_input_labels=None, num_nominal_output_labels=None):
         """
         Initializes the estimator. Can be either instantiated via the following priority of parameters:
         1. JB_Object representing a Java Filter object
@@ -29,6 +30,10 @@ def __init__(self, jobject=None, filter=None, classname=None, options=None):
         :type classname: str
         :param options: the command-line options of the Weka filter to instantiate
         :type options: list
+        :param num_nominal_input_labels: the dictionary with the number of labels for the nominal input variables (key is 0-based attribute index)
+        :type num_nominal_input_labels: dict
+        :param num_nominal_output_labels: the number of labels for the output variable
+        :type num_nominal_output_labels: int
         """
         if jobject is not None:
             _jobject = jobject
@@ -51,6 +56,8 @@ def __init__(self, jobject=None, filter=None, classname=None, options=None):
         # the following references are required for get_params/set_params
         self._classname = classname
         self._options = options
+        self._num_nominal_input_labels = num_nominal_input_labels
+        self._num_nominal_output_labels = num_nominal_output_labels
 
     @property
     def filter(self):
@@ -87,7 +94,9 @@ def fit(self, data, targets):
             check_array(data)
         else:
             check_X_y(data, targets)
-        d = to_instances(data, y=targets)
+        d = to_instances(data, y=targets,
+                         num_nominal_labels=self._num_nominal_input_labels,
+                         num_class_labels=self._num_nominal_output_labels)
         self.header_ = Instances.template_instances(d)
         self._filter.inputformat(d)
         self._filter.filter(d)
@@ -114,7 +123,9 @@ def transform(self, data, targets=None):
                 targets.append(missing_value())
             targets = np.array(targets)
 
-        d = to_instances(data, y=targets)
+        d = to_instances(data, y=targets,
+                         num_nominal_labels=self._num_nominal_input_labels,
+                         num_class_labels=self._num_nominal_output_labels)
         d_new = self._filter.filter(d)
         X, y = to_array(d_new)
         if no_targets:
@@ -134,6 +145,10 @@ def get_params(self, deep=True):
         result = dict()
         result["classname"] = self._classname
         result["options"] = self._options
+        if self._num_nominal_input_labels is not None:
+            result["num_nominal_input_labels"] = self._num_nominal_input_labels
+        if self._num_nominal_output_labels is not None:
+            result["num_nominal_output_labels"] = self._num_nominal_output_labels
         return result
 
     def set_params(self, **params):
@@ -152,6 +167,12 @@ def set_params(self, **params):
         self._classname = params["classname"]
         self._options = params["options"]
         self._filter = Filter(classname=self._classname, options=self._options)
+        self._num_nominal_input_labels = None
+        if "num_nominal_input_labels" in params:
+            self._num_nominal_input_labels = params["num_nominal_input_labels"]
+        self._num_nominal_output_labels = None
+        if "num_nominal_output_labels" in params:
+            self._num_nominal_output_labels = params["num_nominal_output_labels"]
 
     def __str__(self):
         """
@@ -194,18 +215,24 @@ class MakeNominal(BaseEstimator, TransformerMixin):
     Converts numeric columns to nominal ones (ie string labels).
     """
 
-    def __init__(self, input_vars=None, output_var=False):
+    def __init__(self, input_vars=None, output_var=False, num_nominal_input_labels=None, num_nominal_output_labels=None):
         """
         Initializes the estimator.
 
         :param nominal_input_vars: the list of 0-based indices of attributes to convert to nominal or range string with 1-based indices
         :type nominal_input_vars: list or str
         :param output_var: whether to convert the output variable as well
         :type output_var: bool
+        :param num_nominal_input_labels: the dictionary with the number of labels for the nominal input variables (key is 0-based attribute index)
+        :type num_nominal_input_labels: dict
+        :param num_nominal_output_labels: the number of labels for the output variable
+        :type num_nominal_output_labels: int
         """
         super(MakeNominal, self).__init__()
         self._input_vars = None if input_vars is None else input_vars[:]
         self._output_var = output_var
+        self._num_nominal_input_labels = num_nominal_input_labels
+        self._num_nominal_output_labels = num_nominal_output_labels
 
     @property
     def input_vars(self):
@@ -275,6 +302,10 @@ def get_params(self, deep=True):
         result = dict()
         result["input_vars"] = self._input_vars
         result["output_var"] = self._output_var
+        if self._num_nominal_input_labels is not None:
+            result["num_nominal_input_labels"] = self._num_nominal_input_labels
+        if self._num_nominal_output_labels is not None:
+            result["num_nominal_output_labels"] = self._num_nominal_output_labels
         return result
 
     def set_params(self, **params):
@@ -292,6 +323,12 @@ def set_params(self, **params):
             raise Exception("Cannot find 'output_var' in parameters!")
         self._input_vars = params["input_vars"]
         self._output_var = params["output_var"]
+        self._num_nominal_input_labels = None
+        if "num_nominal_input_labels" in params:
+            self._num_nominal_input_labels = params["num_nominal_input_labels"]
+        self._num_nominal_output_labels = None
+        if "num_nominal_output_labels" in params:
+            self._num_nominal_output_labels = params["num_nominal_output_labels"]
 
     def __str__(self):
         """