File size: 7,504 Bytes
3d97c90
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
import numpy as np
import math


class DecisionTree:

    class Node:
        def __init__(self, feature=None, threshold=None, left=None, right=None, decision=None):
            self.feature = feature
            self.threshold = threshold
            self.left = left
            self.right = right
            self.decision = decision


    def __init__(self,
                 maxDepth=None,
                 minSamplesSplit=2,
                 criterion="gini",
                 maxFeatures=None,
                 minSampleLeafs=1,
                 classWeights=None,
                 randomState=None):

        self.maxDepth = maxDepth
        self.minSamplesSplit = minSamplesSplit
        self.minSampleLeafs = minSampleLeafs
        self.criterion = criterion
        self.maxFeatures = maxFeatures
        self.allowedCriteria = ["entropy", "gini"]
        self.classWeights = classWeights

        # NEW: deterministic randomness
        self.randomState = randomState
        self.rng = np.random.default_rng(randomState)

        if criterion not in self.allowedCriteria:
            raise ValueError("Criterion has to be either entropy or gini")

        self.root = None


    def majorityLabel(self, labels):
        counts = np.bincount(labels)
        return np.argmax(counts)


    def getThreshold(self, featureValues, labels):

        entropy = []
        labelCount = len(labels)

        sortedIndices = np.argsort(featureValues)
        sortedFeatureValues = featureValues[sortedIndices]
        sortedLabels = labels[sortedIndices]

        leftLabelCounts = np.zeros(np.max(labels) + 1, dtype=int)
        rightLabelCounts = np.bincount(sortedLabels,
                                      minlength=np.max(labels) + 1)

        thresholds = []

        for i in range(len(sortedLabels) - 1):

            leftLabelCounts[sortedLabels[i]] += 1
            rightLabelCounts[sortedLabels[i]] -= 1

            if np.sum(leftLabelCounts) < self.minSampleLeafs:
                continue

            if np.sum(rightLabelCounts) < self.minSampleLeafs:
                continue

            if sortedFeatureValues[i] == sortedFeatureValues[i + 1]:
                continue

            w_entropy = self.computeImpurity(
                leftLabelCounts,
                rightLabelCounts,
                labelCount
            )

            threshold = (
                sortedFeatureValues[i] +
                sortedFeatureValues[i + 1]
            ) / 2

            thresholds.append(threshold)
            entropy.append(w_entropy)

        if len(entropy) == 0:
            return None, None

        minEntropyIndex = np.argmin(entropy)

        return thresholds[minEntropyIndex], entropy[minEntropyIndex]


    def computeImpurity(self,
                        leftLabelCounts,
                        rightLabelCounts,
                        labelCount):

        leftEntropy = 0
        rightEntropy = 0

        if self.classWeights is not None:

            weightsArr = np.array([
                self.classWeights.get(i, 1)
                for i in range(leftLabelCounts.size)
            ])

        else:

            weightsArr = np.ones(leftLabelCounts.size)

        weightedCountsL = weightsArr * leftLabelCounts
        weightedCountsR = weightsArr * rightLabelCounts

        wLeftLabelCount = np.sum(weightedCountsL)
        wRightLabelCount = np.sum(weightedCountsR)

        totalWeight = wLeftLabelCount + wRightLabelCount


        for count in weightedCountsL:

            if count == 0:
                continue

            prob = count / wLeftLabelCount

            if self.criterion == "entropy":
                leftEntropy += -prob * np.log2(prob)
            else:
                leftEntropy += prob ** 2


        for count in weightedCountsR:

            if count == 0:
                continue

            prob = count / wRightLabelCount

            if self.criterion == "entropy":
                rightEntropy += -prob * np.log2(prob)
            else:
                rightEntropy += prob ** 2


        if self.criterion == "entropy":

            w_entropy = (
                (wLeftLabelCount / totalWeight) * leftEntropy +
                (wRightLabelCount / totalWeight) * rightEntropy
            )

        else:

            w_entropy = (
                (wLeftLabelCount / totalWeight) * (1 - leftEntropy) +
                (wRightLabelCount / totalWeight) * (1 - rightEntropy)
            )

        return w_entropy


    def getFeature(self, trainingData, labels):

        entropy = []
        thresholds = []
        featureIndices = []

        noOfFeatures = trainingData.shape[1]


        if self.maxFeatures is not None:

            current_max = self.maxFeatures

            if current_max == "sqrt":
                current_max = int(math.sqrt(noOfFeatures))

            elif current_max == "log2":
                current_max = int(np.log2(noOfFeatures))


            featureArr = self.rng.choice(
                noOfFeatures,
                size=current_max,
                replace=False
            )

        else:

            featureArr = np.arange(noOfFeatures)


        for j in featureArr:

            featureValues = trainingData[:, j]

            optimalThreshold, minEntropy = self.getThreshold(
                featureValues,
                labels
            )

            if optimalThreshold is None:
                continue

            entropy.append(minEntropy)
            thresholds.append(optimalThreshold)
            featureIndices.append(j)


        if len(entropy) == 0:
            return None, None


        minEntropyIndex = np.argmin(entropy)

        return (
            featureIndices[minEntropyIndex],
            thresholds[minEntropyIndex]
        )


    def buildTree(self, data, labels, depth=0):

        if len(np.unique(labels)) == 1:
            return self.Node(decision=np.unique(labels)[0])

        if self.maxDepth is not None and depth >= self.maxDepth:
            return self.Node(decision=self.majorityLabel(labels))

        if len(labels) < self.minSamplesSplit:
            return self.Node(decision=self.majorityLabel(labels))


        feature, threshold = self.getFeature(data, labels)

        if feature is None:
            return self.Node(decision=self.majorityLabel(labels))


        leftMask = data[:, feature] <= threshold
        rightMask = ~leftMask


        leftChild = self.buildTree(
            data[leftMask],
            labels[leftMask],
            depth + 1
        )

        rightChild = self.buildTree(
            data[rightMask],
            labels[rightMask],
            depth + 1
        )


        return self.Node(
            feature,
            threshold,
            leftChild,
            rightChild
        )


    def fit(self, trainingData, labels):

        self.root = self.buildTree(trainingData, labels)


    def traverseTree(self, dataPoint, node):

        if node.decision is not None:
            return node.decision

        if dataPoint[node.feature] <= node.threshold:
            return self.traverseTree(dataPoint, node.left)

        return self.traverseTree(dataPoint, node.right)


    def predict(self, datapoints):

        if self.root is None:
            raise ValueError(
                "Decision tree has not been trained. Call fit() first."
            )

        return np.array([
            self.traverseTree(point, self.root)
            for point in datapoints
        ])