Method classify

pattern/vector/__init__.py:2007–2044

Returns the type with the highest probability for the given document. If the classifier has been trained on LSA concept vectors you need to supply LSA.transform(document).

(self, document, discrete=True)

            self._likelihood[type][f] = self._likelihood[type].get(f, 0) + w

    def classify(self, document, discrete=True):
        """ Returns the type with the highest probability for the given document.
            If the classifier has been trained on LSA concept vectors
            you need to supply LSA.transform(document).
        """
        # Given red & round, what is the likelihood that it is an apple?
        # p = p(red|apple) * p(round|apple) * p(apple) / (p(red) * p(round))
        # The multiplication can cause underflow so we use log() instead.
        # For unknown features, we smoothen with an alpha value.
        v = self._vector(document)[1]
        n = float(sum(self._classes.itervalues()))
        p = defaultdict(float)
        for type in self._classes:
            if self._method == MULTINOMIAL:
                d = float(sum(self._likelihood[type].itervalues()))
            if self._method == BERNOUILLI:
                d = float(self._classes[type])
            g = 0
            for f in v:
                # Conditional probabilities:
                g += log(self._likelihood[type].get(f, self._alpha) / d)
            g = exp(g) * self._classes[type] / n # prior
            p[type] = g
        # Normalize probability estimates.
        s = sum(p.itervalues()) or 1
        for type in p:
            p[type] /= s
        if not discrete:
            return p
        try:
            # Ties are broken in favor of the majority class
            # (random winner for majority ties).
            m = max(p.itervalues())
            p = sorted((self._classes[type], type) for type, g in p.iteritems() if g == m > 0)
            p = [type for frequency, type in p if frequency == p[0][0]]
            return choice(p)
        except:
            return self.baseline

Bayes = NB = NaiveBayes