Class Vector

pattern/vector/__init__.py:612–657

# sum(len(d.vector) for d in model.documents) / float(len(model))

class Vector(readonlydict):
    
    id = 1
    
    def __init__(self, *args, **kwargs):
        """ A dictionary of (feature, weight)-items of the features (terms, words) in a Document.
            A vector can be used to compare the document to another document with a distance metric.
            For example, vectors with 2 features (x, y) can be compared using 2D Euclidean distance.
            Vectors that represent text documents can be compared using cosine similarity.
        """
        self.id     = Vector.id; Vector.id+=1  # Unique ID.
        self.weight = kwargs.pop("weight", TF) # Vector weights based on tf, tf-idf, binary?
        self._norm  = None
        readonlydict.__init__(self, *args, **kwargs)
    
    @property
    def features(self):
        return self.keys()
    
    @property
    def l2_norm(self):
        """ Yields the Frobenius matrix norm (cached).
            n = the square root of the sum of the absolute squares of the values.
            The matrix norm is used to normalize (0.0-1.0) cosine similarity between documents.
        """
        if self._norm is None: 
            self._norm = sum(w * w for w in self.itervalues()) ** 0.5
        return self._norm
        
    norm = l2 = L2 = L2norm = l2norm = L2_norm = l2_norm
    
    def copy(self):
        return Vector(self, weight=self.weight)

    def __call__(self, vector={}):
        """ Vector(vector) returns a new vector updated with values from the given vector.
            No new features are added. For example: Vector({1:1, 2:2})({1:0, 3:3}) => {1:0, 2:2}.
        """
        if isinstance(vector, (Document, Model)):
            vector = vector.vector
        v = self.copy()
        s = dict.__setitem__
        for f, w in vector.iteritems():
            if f in v:
                s(v, f, w)
        return v

#--- VECTOR DISTANCE -------------------------------------------------------------------------------
# The "distance" between two vectors can be calculated using different metrics.