Source Code for Module mekano.ml.multiclassifier

 1  from __future__ import with_statement 
 2  from ..atoms import AtomVector 
 3   
 4 -class MultiClassifier: 
 5      """ 
 6      mc = MultiClassifier() 
 7       
 8      Manages a set of classifiers for a multi-class/multi-label 
 9      classification problem. 
10      There are two ways to create a MultiClassifier. 
11      First, manually create the classifier for each label, then 
12      add it to MultiClasifier: 
13       
14          >>> mc.add(label, classifier) 
15       
16      Each classifier should provide a score function. 
17   
18      scores = mc.score(av) 
19      'scores' is a map of label:score 
20       
21      Second, provide a BaseClassifier class and a multi-labeled dataset 
22      and optimal params to be passed to BaseClassifier, in which case 
23      MultiClassifier with binarize the dataset and do the training: 
24       
25          >>> mc = mekano.MultiClassifier.create(mekano.LogisticRegressionClassifier, trainset, LAMBDA=0.1, c=1.0) 
26       
27      """ 
28   
29 -    def __init__(self): 
30          self.classifiers = {} 
31          self.labelset = set() 
32       
33 -    def add(self, label, classifier): 
34          self.classifiers[label] = classifier 
35          self.labelset.add(label) 
36       
37 -    def __getitem__(self, key): 
38          return self.classifiers[key] 
39       
40 -    def score(self, av): 
41          """Score a vector av 
42           
43          Returns a map of label:score 
44          """ 
45          return dict([(l,c.score(av)) for l,c in self.classifiers.iteritems()]) 
46       
47 -    def __repr__(self): 
48          return "<MultiClassfier: %d classifiers>" % (len(self.classifiers)) 
49       
50      @staticmethod 
51 -    def create(BaseClassifier, ds, **params): 
52          """Create a MultiClassifier from a base classifier class and multi-labeled dataset. 
53           
54          'params' are optional parameters to pass to the BaseClassifier constructor. 
55          """ 
56          ret = MultiClassifier() 
57          bds = ds.binarize() 
58          for label, bd in bds.iteritems(): 
59              print "MultiClassifier: Training for", label 
60              baseClassifier = BaseClassifier(**params) 
61              baseClassifier.train(bd) 
62              ret.add(label, baseClassifier) 
63          return ret 
64