Пример #1
0
    def __init__(self, angles, axes_order, name=None):
        if len(angles) != len(axes_order):
            raise InputParameterError(
                "Number of angles must equal number of axes in axes_order.")

        self.axes = ['x', 'y', 'z']
        unrecognized = set(axes_order).difference(self.axes)
        if unrecognized:
            raise ValueError("Unrecognized axis label {0}; "
                             "should be one of {1} ".format(
                                 unrecognized, self.axes))
        self.axes_order = axes_order

        self._func_map = {'x': self._xrot, 'y': self._yrot, 'z': self._zrot}
        super(Rotation3DToGWA, self).__init__(angles, name=name)
Пример #2
0
    def __init__(self, amplitude=amplitude.default, x_mean=x_mean.default,
                 y_mean=y_mean.default, x_stddev=None, y_stddev=None,
                 theta=None, cov_matrix=None, **kwargs):
        if cov_matrix is None:
            if x_stddev is None:
                x_stddev = self.__class__.x_stddev.default
            if y_stddev is None:
                y_stddev = self.__class__.y_stddev.default
            if theta is None:
                theta = self.__class__.theta.default
        else:
            if x_stddev is not None or y_stddev is not None or theta is not None:
                raise InputParameterError("Cannot specify both cov_matrix and "
                                          "x/y_stddev/theta")
            else:
                # Compute principle coordinate system transformation
                cov_matrix = np.array(cov_matrix)

                if cov_matrix.shape != (2, 2):
                    # TODO: Maybe it should be possible for the covariance matrix
                    # to be some (x, y, ..., z, 2, 2) array to be broadcast with
                    # other parameters of shape (x, y, ..., z)
                    # But that's maybe a special case to work out if/when needed
                    raise ValueError("Covariance matrix must be 2x2")

                eig_vals, eig_vecs = np.linalg.eig(cov_matrix)
                x_stddev, y_stddev = np.sqrt(eig_vals)
                y_vec = eig_vecs[:, 0]
                theta = np.arctan2(y_vec[1], y_vec[0])

        # Ensure stddev makes sense if its bounds are not explicitly set.
        # stddev must be non-zero and positive.
        # TODO: Investigate why setting this in Parameter above causes
        #       convolution tests to hang.
        kwargs.setdefault('bounds', {})
        kwargs['bounds'].setdefault('x_stddev', (FLOAT_EPSILON, None))
        kwargs['bounds'].setdefault('y_stddev', (FLOAT_EPSILON, None))

        super().__init__(
            amplitude=amplitude, x_mean=x_mean, y_mean=y_mean,
            x_stddev=x_stddev, y_stddev=y_stddev, theta=theta, **kwargs)
Пример #3
0
 def _compute_matrix(angles, axes_order):
     if len(angles) != len(axes_order):
         raise InputParameterError(
             "Number of angles must equal number of axes in axes_order.")
     matrices = []
     for angle, axis in zip(angles, axes_order):
         matrix = np.zeros((3, 3), dtype=np.float)
         if axis == 'x':
             mat = Rotation3D.rotation_matrix_from_angle(angle)
             matrix[0, 0] = 1
             matrix[1:, 1:] = mat
         elif axis == 'y':
             mat = Rotation3D.rotation_matrix_from_angle(-angle)
             matrix[1, 1] = 1
             matrix[0, 0] = mat[0, 0]
             matrix[0, 2] = mat[0, 1]
             matrix[2, 0] = mat[1, 0]
             matrix[2, 2] = mat[1, 1]
         elif axis == 'z':
             mat = Rotation3D.rotation_matrix_from_angle(angle)
             matrix[2, 2] = 1
             matrix[:2, :2] = mat
         else:
             raise ValueError("Expected axes_order to be a combination \
                     of characters 'x', 'y' and 'z', got {0}".format(
                 set(axes_order).difference(['x', 'y', 'z'])))
         matrices.append(matrix)
     if len(angles) == 1:
         return matrix
     elif len(matrices) == 2:
         return np.dot(matrices[1], matrices[0])
     else:
         prod = np.dot(matrices[1], matrices[0])
         for m in matrices[2:]:
             prod = np.dot(m, prod)
         return prod