refactor Discretization in datasets

benchmark/Datasets.py

@@ -3,7 +3,7 @@ import pandas as pd
 from scipy.io import arff
 from .Utils import Files
 from .Arguments import EnvData
-from mdlp import MDLP
+from mdlp.discretization import MDLP
 
 
 class Diterator:
@@ -93,6 +93,11 @@ class Datasets:
             __import__(__name__),
             f"Datasets{envData['source_data']}",
         )
+        self.load = (
+            self.load_discretized
+            if envData["discretize"] == "1"
+            else self.load_continuous
+        )
         self.dataset = class_name()
         self.class_names = []
         self._load_names()
@@ -128,28 +133,13 @@ class Datasets:
     def get_class_name(self):
         return self.dataset.class_name
 
-    def load(self, name, dataframe=False):
+    def load_continuous(self, name, dataframe=False):
         try:
             class_name = self.class_names[self.data_sets.index(name)]
             return self.dataset.load(name, class_name, dataframe)
         except (ValueError, FileNotFoundError):
             raise ValueError(f"Unknown dataset: {name}")
 
-    def __iter__(self) -> Diterator:
-        return Diterator(self.data_sets)
-
-
-class Discretizer(Datasets):
-    def __init__(self, dataset_name=None):
-        super().__init__(dataset_name)
-
-    def load(self, name, dataframe=False):
-        X, y = super().load(name)
-        X, y = self.discretize(X, y)
-        dataset = pd.DataFrame(X, columns=self.get_features())
-        dataset[self.get_class_name()] = y
-        return dataset if dataframe else X, y
-
     def discretize(self, X, y):
         """Supervised discretization with Fayyad and Irani's MDLP algorithm.
 
@@ -167,3 +157,13 @@ class Discretizer(Datasets):
         discretiz = MDLP()
         Xdisc = discretiz.fit_transform(X, y)
         return Xdisc.astype(int), y.astype(int)
+
+    def load_discretized(self, name, dataframe=False):
+        X, y = self.load_continuous(name)
+        X, y = self.discretize(X, y)
+        dataset = pd.DataFrame(X, columns=self.get_features())
+        dataset[self.get_class_name()] = y
+        return dataset if dataframe else X, y
+
+    def __iter__(self) -> Diterator:
+        return Diterator(self.data_sets)

benchmark/Models.py

@@ -8,6 +8,7 @@ from sklearn.ensemble import (
 )
 from sklearn.svm import SVC
 from stree import Stree
+from bayesclass import TAN
 from wodt import Wodt
 from odte import Odte
 from xgboost import XGBClassifier
@@ -20,6 +21,7 @@ class Models:
     def define_models(random_state):
         return {
             "STree": Stree(random_state=random_state),
+            "TAN": TAN(random_state=random_state),
             "Cart": DecisionTreeClassifier(random_state=random_state),
             "ExtraTree": ExtraTreeClassifier(random_state=random_state),
             "Wodt": Wodt(random_state=random_state),

benchmark/__init__.py

@@ -3,7 +3,6 @@ from .Datasets import (
     DatasetsSurcov,
     DatasetsTanveer,
     DatasetsArff,
-    Discretizer,
 )
 from .Experiments import Experiment
 from .Results import Report, Summary
@@ -13,4 +12,4 @@ __copyright__ = "Copyright 2020-2022, Ricardo Montañana Gómez"
 __license__ = "MIT License"
 __author_email__ = "ricardo.montanana@alu.uclm.es"
 
-__all__ = ["Experiment", "Datasets", "Report", "Summary", "Discretizer"]
+__all__ = ["Experiment", "Datasets", "Report", "Summary"]

benchmark/tests/.env.arff

@@ -4,4 +4,5 @@ n_folds=5
 model=ODTE
 stratified=0
 source_data=Arff
-seeds=[271, 314, 171]
+seeds=[271, 314, 171]
+discretize=1

benchmark/tests/.env.dist

@@ -6,3 +6,4 @@ stratified=0
 # Source of data Tanveer/Surcov
 source_data=Tanveer
 seeds=[57, 31, 1714, 17, 23, 79, 83, 97, 7, 1]
+discretize=0

benchmark/tests/.env.surcov

@@ -5,4 +5,5 @@ model=ODTE
 stratified=0
 # Source of data Tanveer/Surcov
 source_data=Surcov
-seeds=[57, 31, 1714, 17, 23, 79, 83, 97, 7, 1]
+seeds=[57, 31, 1714, 17, 23, 79, 83, 97, 7, 1]
+discretize=0

requirements.txt

@@ -2,7 +2,7 @@ pandas
 scikit-learn
 scipy
 odte
-mdlp
+mdlp-discretization
 mufs
 xlsxwriter
 openpyxl