FD step size on individual design variables

Question

I'm a little bit confused about how to specify FD step sizes on the individual design variables. In the old OpenMDAO it was possible to specify the step size when adding the desvar. In the new version add_desvar does not have this argument, so I tried to set it on the IndepVarComp's fd_options, which didn't work either. It is not so intuitive to me that I have to set it on the downstream component. But in case I had to do it this way, how would I set it on each individual param at setup time? I'm probably missing something, but could you please let me know how to to it on the below example where I'm trying to set the step size of prob.root.sub.p1 with force_fd on the whole problem.

from openmdao.api import Component, Group, Problem, IndepVarComp

class MyComp(Component):

    def __init__(self):
        super(MyComp, self).__init__()

        # Params
        self.add_param('x1', 3.0, step_size = 1e-2)
        self.add_param('x2', 3.0)

        # Unknowns
        self.add_output('y', 5.5)

    def solve_nonlinear(self, params, unknowns, resids):
        """ Doesn't do much. """
        unknowns['y'] = 7.0*params['x1']**2 + 7.0*params['x2']**2

prob = Problem()
prob.root = Group()
sub = prob.root.add('sub', Group())
comp = prob.root.add('comp', MyComp())
p1 = sub.add('p1', IndepVarComp([('x1', 3.0), ('x2', 3.0)]))

# this has no effect
prob.root.sub.p1.fd_options['step_size'] = 1.0e-1



prob.root.connect('sub.p1.x1', 'comp.x1')
prob.root.connect('sub.p1.x2', 'comp.x2')

prob.root.fd_options['force_fd'] = True 

prob.setup(check=False)
prob.run()

J = prob.calc_gradient(['sub.p1.x1', 'sub.p1.x2'], ['comp.y'], return_format='dict')

print J

If I print mydict in System.fd_jacobian and the step size I see:

('mydict', {'shape': 1, 'step_size': 0.01, 'pathname': 'comp.x1', 'val': 3.0, 'promoted_name': 'comp.x1', 'top_promoted_name': 'comp.x1', 'size': 1})
('FD', 'comp.x1', 0.01)
('mydict', {'val': 3.0, 'promoted_name': 'comp.x2', 'shape': 1, 'pathname': 'comp.x2', 'top_promoted_name': 'comp.x2', 'size': 1})
('FD', 'comp.x2', 1e-06)

{'comp.y': {'sub.p1.x1': array([[ 42.07]]), 'sub.p1.x2': array([[ 42.000007]])}}

Thanks!

score 2 · Answer 1 · answered Nov 06 '15 at 19:57

You are right though, we used to also have it on the old add_parametercall (the equivalent of add_desvar now) in classic OpenMDAO, and we haven't re-implemented that yet, so it looks like we still need to do this.

In the mean time, the only way to control stepsize at the var level is to set step_size in the downstream target of the IndepVarComp of the connection, which you do with x1 in your example. It's a workaround that will accomplish what you want, but it is a little nonintuitive and possibly tedious for broadcast connections unfortunately.

FD step size on individual design variables

1 Answers1