Source Code for Module ml.var.svmeasy

  1  ''' 
  2   
  3  @author: Eleftherios Avramidis 
  4  ''' 
  5   
  6  from Orange.classification.svm import SVMLearner, SVMClassifierWrapper 
  7  from Orange.optimization import TuneMParameters 
  8  from Orange.core import DomainContinuizer 
  9   
 10   
 11 -class SVMEasyLearner(SVMLearner): 
 12      ''' 
 13      classdocs 
 14      ''' 
 15 -    def __init__(self, **kwds): 
 16          self.folds=4 
 17          self.verbose=True 
 18          SVMLearner.__init__(self, **kwds) 
 19          self.learner = SVMLearner(**kwds) 
 20          try:  
 21              self.multinomialTreatment = eval(kwds["multinomialTreatment"]) 
 22          except: 
 23              self.multinomialTreatment = DomainContinuizer.NValues 
 24          try: 
 25              self.continuousTreatment = eval(kwds["continuousTreatment"]) 
 26          except: 
 27              self.continuousTreatment = DomainContinuizer.NormalizeBySpan 
 28          try:             
 29              self.classTreatment = eval(kwds["classTreatment"]) 
 30          except: 
 31              self.classTreatment = DomainContinuizer.Ignore 
 32   
 33 -    def __call__(self, examples, weight=0):         
 34          print "continuizing data" 
 35          newdomain, newexamples = self.continuize(examples) 
 36          newexamples.save("continuized_examples.tab") 
 37          self.verbose = True 
 38          print "tuning classifier" 
 39          classifier, fittedParameters = self.tuneClassifier(newexamples, newdomain, examples) 
 40           
 41          paramfile = open("svm.params", 'w') 
 42          print fittedParameters 
 43          for param in fittedParameters: 
 44              paramfile.write(str(param)) 
 45              paramfile.write("\n") 
 46           
 47          paramfile.close() 
 48          return classifier 
 49           
 50 -    def continuize(self, examples): 
 51          transformer=DomainContinuizer() 
 52          #transformer.multinomialTreatment=orange.DomainContinuizer.NValues 
 53          #transformer.continuousTreatment=orange.DomainContinuizer.NormalizeBySpan 
 54          transformer.multinomialTreatment = self.multinomialTreatment 
 55          transformer.continuousTreatment = self.continuousTreatment 
 56          transformer.classTreatment = self.classTreatment 
 57          #transformer.classTreatment = orange.DomainContinuizer.Ignore 
 58          newdomain = transformer(examples) 
 59          newexamples = examples.translate(newdomain) 
 60          return newdomain, newexamples 
 61          #print newexamples[0] 
 62           
 63 -    def tuneClassifier(self, newexamples, newdomain, examples): 
 64          params={} 
 65          parameters = [] 
 66          self.learner.normalization = False ## Normalization already done 
 67           
 68          if self.svm_type in [1,4]: 
 69              numOfNuValues=9 
 70              if self.svm_type == SVMLearner.Nu_SVC: 
 71                  maxNu = max(self.maxNu(newexamples) - 1e-7, 0.0) 
 72              else: 
 73                  maxNu = 1.0 
 74              parameters.append(("nu", [i/10.0 for i in range(1, 9) if i/10.0 < maxNu] + [maxNu])) 
 75          else: 
 76              parameters.append(("C", [2**a for a in  range(-5,15,2)])) 
 77          if self.kernel_type==2: 
 78              parameters.append(("gamma", [2**a for a in range(-5,5,2)]+[0])) 
 79          tunedLearner = TuneMParameters(object=self.learner, parameters=parameters, folds=self.folds) 
 80          print "apply learner" 
 81          import orngWrap 
 82          tunedLearner = orngWrap.TuneMParameters(object=self.learner, 
 83                                                  parameters=parameters, 
 84                                                  folds=self.folds) 
 85   
 86                
 87          return SVMClassifierWrapper(tunedLearner(newexamples, 
 88                                                   verbose=self.verbose)) 
 89   
 90  #        return appliedTunedLearner, appliedTunedLearner.fittedParameters 
 91   
 92   
 93               
 94           
 95  #examples = orange.ExampleTable("/home/Eleftherios Avramidis/workspace/TaraXUscripts/src/training-attset1.100.tab") 
 96  #svmlearner = SVMEasy(examples) 
 97  # 
 98  #testexamples = orange.ExampleTable("/home/Eleftherios Avramidis/workspace/TaraXUscripts/src/training-attset2.tab") 
 99  #print svmlearner(testexamples[1]) 
100   
101  #objectfile = open("/tmp/svmeasy.pickle", 'w') 
102  #pickle.dump(svmlearner, objectfile) 
103  #objectfile.close() 
104  # 
105  #objectfile = open("/tmp/svmeasy.pickle", 'r') 
106  #recovered_svmlearner = pickle.load(objectfile) 
107  #objectfile.close() 
108  #print recovered_svmlearner(testexamples[1]) 
109