serving.py 文件源码

def latent_correlation(self):
        """Compute correlation matrix among latent features.

        This computes the generalization of Pearson's correlation to discrete
        data. Let I(X;Y) be the mutual information. Then define correlation as

          rho(X,Y) = sqrt(1 - exp(-2 I(X;Y)))

        Returns:
            A [V, V]-shaped numpy array of feature-feature correlations.
        """
        logger.debug('computing latent correlation')
        V, E, M, R = self._VEMR
        edge_probs = self._edge_probs
        vert_probs = self._vert_probs
        result = np.zeros([V, V], np.float32)
        for root in range(V):
            messages = np.empty([V, M, M])
            program = make_propagation_program(self._tree.tree_grid, root)
            for op, v, v2, e in program:
                if op == OP_ROOT:
                    # Initialize correlation at this node.
                    messages[v, :, :] = np.diagflat(vert_probs[v, :])
                elif op == OP_OUT:
                    # Propagate correlation outward from parent to v.
                    trans = edge_probs[e, :, :]
                    if v > v2:
                        trans = trans.T
                    messages[v, :, :] = np.dot(  #
                        trans / vert_probs[v2, np.newaxis, :],
                        messages[v2, :, :])
            for v in range(V):
                result[root, v] = correlation(messages[v, :, :])
        return result