def compare_jacobian(self, fit_function, ndim, isotropic, n, groups=None,
custom_param_mode=None):
ff = FitFunctions(fit_function, ndim, isotropic)
param_mode = {param: 'var' for param in ff.params}
param_mode['background'] = 'cluster'
if custom_param_mode is not None:
param_mode.update(custom_param_mode)
ff = FitFunctions(fit_function, ndim, isotropic, param_mode=param_mode)
params = np.random.random((n, len(ff.params))) * 10
residual, jacobian = self.get_residual(ff, n, params, groups)
vect = vect_from_params(params, ff.modes, groups, operation=np.mean)
actual = jacobian(vect)
expected = approx_jacobian(vect, residual, EPSILON)[0]
assert_allclose(actual, expected, rtol=RTOL, atol=ATOL)
def test_custom_no_jac(self):
fit_function = dict(name='parabola', params=['a'],
fun=lambda r2, p: p[0]*r2)
ff = FitFunctions(fit_function, ndim=2, isotropic=True)
params = np.random.random((1, len(ff.params))) * 10
residual, jacobian = self.get_residual(ff, 1, params)
assert jacobian is None
def test_2D_gauss(self):
def gauss(p, y, x):
background, signal, yc, xc, size = p
return background + signal*np.exp(-((y - yc)**2/size**2 + (x - xc)**2/size**2))
ff = FitFunctions('gauss', ndim=2, isotropic=True)
params = np.array([[5, 200, 4, 5, 6]], dtype=np.float)
image = np.random.random(self.repeats) * 200
mesh = np.random.random((ff.ndim, self.repeats)) * 10
masks = np.full((1, self.repeats), True, dtype=np.bool)
residual, jacobian = ff.get_residual([image], [mesh], [masks], params)
vect = vect_from_params(params, ff.modes, groups=None, operation=np.mean)
actual = residual(vect)
expected = np.sum((image - gauss(params[0], *mesh))**2) / self.repeats
assert_allclose(actual, expected, atol=1E-7)
def test_2D_gauss(self):
def gauss(p, y, x):
background, signal, yc, xc, size = p
return background + signal*np.exp(-((y - yc)**2/size**2 + (x - xc)**2/size**2))
ff = FitFunctions('gauss', ndim=2, isotropic=True)
params = np.array([[5, 200, 4, 5, 6]], dtype=np.float)
image = np.random.random(self.repeats) * 200
mesh = np.random.random((ff.ndim, self.repeats)) * 10
masks = np.full((1, self.repeats), True, dtype=np.bool)
residual, jacobian = ff.get_residual([image], [mesh], [masks], params)
vect = vect_from_params(params, ff.modes, groups=None, operation=np.mean)
actual = residual(vect)
expected = np.sum((image - gauss(params[0], *mesh))**2) / self.repeats
assert_allclose(actual, expected, atol=1E-7)