Class ConfusionMatrix

pattern/vector/__init__.py:1862–1909

#--- CLASSIFIER EVALUATION -------------------------------------------------------------------------

class ConfusionMatrix(defaultdict):
    
    def __init__(self, classify=lambda document: True, documents=[]):
        """ Returns the matrix of classes x predicted classes as a dictionary.
        """
        defaultdict.__init__(self, lambda: defaultdict(int))
        for document, type1 in documents:
            type2 = classify(document)
            self[type1][type2] += 1

    def split(self):
        """ Returns an iterator over one-vs-all (type, TP, TN, FP, FN)-tuples.
        """
        return iter((type,) + self(type) for type in self)

    def __call__(self, type):
        """ Returns a (TP, TN, FP, FN)-tuple for the given class (one-vs-all).
        """
        TP = 0 # True positives.
        TN = 0 # True negatives.
        FP = 0 # False positives (type I error).
        FN = 0 # False negatives (type II error).
        for t1 in self:
            for t2, n in self[t1].iteritems():
                if type == t1 == t2: 
                    TP += n
                if type != t1 == t2: 
                    TN += n
                if type == t1 != t2: 
                    FN += n
                if type == t2 != t1: 
                    FP += n
        return (TP, TN, FP, FN)
    
    @property
    def table(self, padding=1):
        k = sorted(self)
        n = max(map(lambda x: len(decode_utf8(x)), k))
        n = max(n, *(len(str(self[k1][k2])) for k1 in k for k2 in k)) + padding
        s = "".ljust(n)
        for t1 in k:
            s += decode_utf8(t1).ljust(n)
        for t1 in k:
            s += "\n"
            s += decode_utf8(t1).ljust(n)
            for t2 in k: 
                s += str(self[t1][t2]).ljust(n)
        return s

def K_fold_cross_validation(Classifier, documents=[], folds=10, **kwargs):
    """ For 10-fold cross-validation, performs 10 separate tests of the classifier,