Source Code for Module rdkit.ML.NaiveBayes.CrossValidate

 1  # $Id: CrossValidate.py 997 2009-02-25 06:12:43Z glandrum $ 
 2  # 
 3  #  Copyright (C) 2004-2005 Rational Discovery LLC. 
 4  #   All Rights Reserved 
 5  # 
 6  """ handles doing cross validation with naive bayes models 
 7  and evaluation of individual models 
 8   
 9  """ 
10  from rdkit.ML.NaiveBayes.ClassificationModel import NaiveBayesClassifier 
11  from rdkit.ML.Data import SplitData 
12  from rdkit.ML.FeatureSelect import CMIM 
13   
14 -def makeNBClassificationModel(trainExamples, attrs, nPossibleValues, nQuantBounds, 
15                                mEstimateVal=-1.0, 
16                                useSigs=False, 
17                                ensemble=None,useCMIM=0, 
18                                **kwargs) : 
19    if useCMIM > 0 and useSigs and not ensemble: 
20      ensemble = CMIM.SelectFeatures(trainExamples,useCMIM,bvCol=1) 
21    if ensemble: 
22      attrs = ensemble 
23    model = NaiveBayesClassifier(attrs, nPossibleValues, nQuantBounds,  
24                                 mEstimateVal=mEstimateVal,useSigs=useSigs) 
25   
26   
27    model.SetTrainingExamples(trainExamples) 
28    model.trainModel() 
29    return model 
30       
31 -def CrossValidate(NBmodel, testExamples, appendExamples=0) : 
32       
33    nTest = len(testExamples) 
34    assert nTest,'no test examples: %s'%str(testExamples) 
35    badExamples = [] 
36    nBad = 0 
37    preds = NBmodel.ClassifyExamples(testExamples, appendExamples) 
38    assert len(preds) == nTest 
39   
40    for i in range(nTest): 
41      testEg = testExamples[i] 
42      trueRes = testEg[-1] 
43      res = preds[i] 
44   
45      if (trueRes != res) : 
46        badExamples.append(testEg) 
47        nBad += 1 
48    return float(nBad)/nTest, badExamples 
49   
50 -def CrossValidationDriver(examples, attrs, nPossibleValues, nQuantBounds, 
51                            mEstimateVal=0.0, 
52                            holdOutFrac=0.3, modelBuilder=makeNBClassificationModel, 
53                            silent=0, calcTotalError=0, **kwargs) : 
54    nTot = len(examples) 
55    if not kwargs.get('replacementSelection',0): 
56      testIndices,trainIndices = SplitData.SplitIndices(nTot,holdOutFrac, 
57                                                        silent=1,legacy=1, 
58                                                        replacement=0) 
59    else : 
60      testIndices,trainIndices = SplitData.SplitIndices(nTot,holdOutFrac, 
61                                                        silent=1,legacy=0, 
62                                                        replacement=1) 
63   
64    trainExamples = [examples[x] for x in trainIndices] 
65    testExamples = [examples[x] for x in testIndices] 
66   
67    NBmodel = modelBuilder(trainExamples, attrs, nPossibleValues, nQuantBounds, 
68                           mEstimateVal,**kwargs) 
69   
70    if not calcTotalError:                #  
71      xValError, badExamples = CrossValidate(NBmodel, testExamples,appendExamples=1) 
72    else: 
73      xValError,badExamples = CrossValidate(NBmodel, examples,appendExamples=0) 
74   
75    if not silent: 
76      print 'Validation error was %%%4.2f'%(100*xValError) 
77    NBmodel._trainIndices = trainIndices 
78    return NBmodel, xValError 
79