Updated model classifier selection + added references to literature #151

Author: dpuenteramirez

semisupervised/CoTraining.py

@@ -3,7 +3,7 @@
 # @Filename:    CoTraining.py
 # @Author:      Daniel Puente Ramírez
 # @Time:        22/12/21 09:27
-# @Version:     3.0
+# @Version:     4.0
 
 from math import ceil, floor
 
@@ -14,16 +14,35 @@
 
 
 class CoTraining:
-
-    def __init__(self, p=1, n=3, k=30, u=75, random_state=None):
+    """Blum, A., & Mitchell, T. (1998, July). Combining labeled and unlabeled
+        data with co-training. In Proceedings of the eleventh annual conference
+        on Computational learning theory (pp. 92-100).
+     """
+
+    def __init__(self, p=1, n=3, k=30, u=75, random_state=None,
+                 c1=None, c1_params=None,
+                 c2=None, c2_params=None,
+                 ):
         self.p = p
         self.n = n
         self.k = k
         self.u = u
         self.random_state = random_state
         self.size_x1 = 0
-        self.h1 = GaussianNB()
-        self.h2 = GaussianNB()
+
+        classifiers = [c1, c2]
+        classifiers_params = [c1_params, c2_params]
+        configs = []
+        for c, cp in zip(classifiers, classifiers_params):
+            if c is not None:
+                if cp is not None:
+                    configs.append(c(**cp))
+                else:
+                    configs.append(c())
+            else:
+                configs.append(GaussianNB())
+
+        self.h1, self.h2 = configs
 
     def fit(self, samples, y):
         labeled, u, y = split(samples, y)

semisupervised/DemocraticCoLearning.py

@@ -3,7 +3,7 @@
 # @Filename:    DemocraticCoLearning.py
 # @Author:      Daniel Puente Ramírez
 # @Time:        29/12/21 15:39
-# @Version:     3.0
+# @Version:     4.0
 
 import copy
 from math import sqrt
@@ -38,7 +38,10 @@ class DemocraticCoLearning:
         Intelligence (pp. 594-602). IEEE.
     """
 
-    def __init__(self, n_neighbors=3, random_state=None):
+    def __init__(self, random_state=None,
+                 c1=None, c1_params=None,
+                 c2=None, c2_params=None,
+                 c3=None, c3_params=None):
         self.const = 1.96  # 95%
         self.random_state = random_state if random_state is not None else \
             np.random.randint(low=0, high=10e5, size=1)[0]
@@ -49,9 +52,21 @@ def __init__(self, n_neighbors=3, random_state=None):
         self.w2 = 0
         self.w3 = 0
 
-        self.h1 = MultinomialNB()
-        self.h2 = KNeighborsClassifier(n_neighbors=n_neighbors, n_jobs=-1, p=2)
-        self.h3 = DecisionTreeClassifier(random_state=self.random_state)
+        classifiers = [c1, c2, c3]
+        classifiers_params = [c1_params, c2_params, c3_params]
+        default_classifiers = [MultinomialNB, KNeighborsClassifier,
+                               DecisionTreeClassifier]
+        configs = []
+        for index, (c, cp) in enumerate(zip(classifiers, classifiers_params)):
+            if c is not None:
+                if cp is not None:
+                    configs.append(c(**cp))
+                else:
+                    configs.append(c())
+            else:
+                configs.append(default_classifiers[index]())
+
+        self.h1, self.h2, self.h3 = configs
 
     def fit(self, samples, y):
         labeled, u, y = split(samples, y)

semisupervised/DensityPeaks.py

@@ -14,9 +14,20 @@
 from sklearn.preprocessing import LabelEncoder
 from sklearn.semi_supervised import SelfTrainingClassifier
 from sklearn.svm import SVC
+from .utils import split
+from instance_selection import ENN
 
 
 class STDPNF:
+    """
+    Li, J., Zhu, Q., & Wu, Q. (2019). A self-training method based on density
+        peaks and an extended parameter-free local noise filter for k nearest
+        neighbor. Knowledge-Based Systems, 184, 104895.
+
+    Wu, D., Shang, M., Luo, X., Xu, J., Yan, H., Deng, W., & Wang, G. (2018).
+     Self-training semi-supervised classification based on density peaks of
+      data. Neurocomputing, 275, 180-191.
+    """
 
     def __init__(self,
                  dc=None,

semisupervised/TriTraining.py

@@ -3,7 +3,7 @@
 # @Filename:    TriTraining.py
 # @Author:      Daniel Puente Ramírez
 # @Time:        27/12/21 10:25
-# @Version:     3.0
+# @Version:     4.0
 
 from math import floor, ceil
 
@@ -27,23 +27,31 @@ def measure_error(classifier_j, classifier_k, labeled_data):
 
 
 class TriTraining:
-    def __init__(self, learn, random_state=None):
-        if learn == '3-NN':
-            self.hj = KNeighborsClassifier(n_neighbors=3, n_jobs=-1, p=2)
-            self.hk = KNeighborsClassifier(n_neighbors=3, n_jobs=-1, p=2)
-            self.hi = KNeighborsClassifier(n_neighbors=3, n_jobs=-1, p=2)
-        elif learn == 'DecisionTree Classifier':
-            self.hj = DecisionTreeClassifier(random_state=random_state)
-            self.hk = DecisionTreeClassifier(random_state=random_state)
-            self.hi = DecisionTreeClassifier(random_state=random_state)
-        elif learn == 'RandomForest Classifier':
-            self.hj = RandomForestClassifier(random_state=random_state)
-            self.hk = RandomForestClassifier(random_state=random_state)
-            self.hi = RandomForestClassifier(random_state=random_state)
-        else:
-            self.hj = GaussianNB()
-            self.hk = GaussianNB()
-            self.hi = GaussianNB()
+    """Zhou, Z. H., & Li, M. (2005). Tri-training: Exploiting unlabeled data
+        using three classifiers. IEEE Transactions on knowledge and Data
+        Engineering, 17(11), 1529-1541.
+    """
+
+    def __init__(self, random_state=None,
+                 c1=None, c1_params=None,
+                 c2=None, c2_params=None,
+                 c3=None, c3_params=None):
+
+        classifiers = [c1, c2, c3]
+        classifiers_params = [c1_params, c2_params, c3_params]
+        default_classifiers = [KNeighborsClassifier, DecisionTreeClassifier,
+                               RandomForestClassifier]
+        configs = []
+        for index, (c, cp) in enumerate(zip(classifiers, classifiers_params)):
+            if c is not None:
+                if cp is not None:
+                    configs.append(c(**cp))
+                else:
+                    configs.append(c())
+            else:
+                configs.append(default_classifiers[index]())
+
+        self.hj, self.hk, self.hi = configs
 
         self.random_state = random_state if random_state is not None else \
             np.random.randint(low=0, high=10e5, size=1)[0]

semisupervised/utils/_split.py

@@ -5,6 +5,7 @@
 # @Time:        4/2/22 11:54
 
 import numpy as np
+import pandas as pd
 
 
 def split(samples, y):
@@ -20,10 +21,13 @@ def split(samples, y):
         y {Numpy array} -- real labels
     """
 
+    if isinstance(y, pd.DataFrame):
+        y = y.to_numpy()
+
     labeled_indexes = y != (-1 or np.NaN or None)
 
-    L = samples[labeled_indexes].to_numpy()
-    U = samples[~labeled_indexes].to_numpy()
+    L = samples.iloc[labeled_indexes].to_numpy()
+    U = samples.iloc[~labeled_indexes].to_numpy()
     y = y[labeled_indexes]
 
     assert len(L) == len(y), f"L {len(L)} != {len(y)} y"