d1ac7b831ad36cd0e4bdd7980819f83208345148,gpflow/expectations.py,,_expectation,#Any#Any#Any#Any#Any#,560

Before Change

    if not kern.on_separate_dimensions:
        raise NotImplementedError("Product currently needs to be defined on separate dimensions.")  // pragma: no cover
    with tf.control_dependencies([
        tf.assert_equal(tf.rank(p.var), 2,
                        message="Product currently only supports diagonal Xcov.", name="assert_Xcov_diag"),
    ]):
        _expectation_fn = lambda k: _expectation(p, k, feat, k, feat)
        return functools.reduce(tf.multiply, [_expectation_fn(k) for k in kern.kern_list])


@dispatch(DiagonalGaussian, object, (InducingFeature, type(None)), object, (InducingFeature, type(None)))
def _expectation(p, obj1, obj2, obj3, obj4):
    gauss = Gaussian(p.mu, tf.matrix_diag(p.var))
    return _expectation(gauss, obj1, obj2, obj3, obj4)

After Change

    :return: NxDxQ
    
    with params_as_tensors_for(mean1), params_as_tensors_for(mean2):
        N = tf.shape(p.mu)[0]
        e_xxt = p.cov + (p.mu[:, :, None] * p.mu[:, None, :])  // NxDxD
        e_xxt_A = tf.matmul(e_xxt, tf.tile(mean2.A[None, ...], (N, 1, 1)))  // NxDxQ
        e_x_bt = p.mu[:, :, None] * mean2.b[None, None, :]  // NxDxQ

Instances