How can I add an integer variable to an openmdao driver?

Question

I am confused by this openmdao error. Why is this being raised? Can I somehow tell openmdao that I don't have gradients and to use finite differences? Why is this raised for childWeight but not eta?

I can get past this problem by initializing al my variables as floating point (eg. root.add('childWeight', IndepVarComp('x',100)) -> root.add('childWeight', IndepVarComp('x',100.0))), but I would like to understand why this error was raised.

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

class gym(Component):
    def __init__(self):
        super(gym, self).__init__()
        self.add_param('eta', 0.01)
        self.add_param('childWeight', 240)

        self.add_output('acc', 1)

    def solve_nonlinear(self, params, unknowns, resids):
        <...... parameters are used to produce objective "acc" ...>
        unknowns["acc"] = ....

top = Problem()
root = top.root = Group()
root.add('gym', gym())
top.driver = ScipyOptimizer()
top.driver.options['optimizer'] = 'BFGS'

root.add('eta',IndepVarComp('x', 0.01))
root.add('childWeight', IndepVarComp('x',100))

root.connect('eta.x', 'gym.eta')
root.connect('childWeight.x', 'gym.childWeight')

top.driver.add_desvar('eta.x', 0, 1.0)
top.driver.add_desvar('childWeight.x', 0, 1000)

top.driver.add_objective('gym.acc')
top.setup()
top.run()

raises the error

  File "script.py", line 98, in <module>
    top.setup()
  File "/usr/local/lib/python2.7/site-packages/openmdao/core/problem.py", line 694, in setup
    self.driver._setup()
  File "/usr/local/lib/python2.7/site-packages/openmdao/drivers/scipy_optimizer.py", line 91, in _setup
    super(ScipyOptimizer, self)._setup()
  File "/usr/local/lib/python2.7/site-packages/openmdao/core/driver.py", line 115, in _setup
    (item_name, name, oname))
RuntimeError: Parameter 'childWeight.x' is a 'pass_by_obj' variable and can't be used with a gradient based driver of type 'BFGS'.

Justin Gray · Accepted Answer · 2016-02-17 21:41:17Z

2

The problem is this line

root.add('childWeight', IndepVarComp('x',240))

You've created an integer variable. Try this instead:

root.add('childWeight', IndepVarComp('x',240.))

If you want to use finite-differences you will also want:

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

answered Feb 17, 2016 at 21:41

Justin Gray

5,7101 gold badge13 silver badges16 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

kilojoules Over a year ago

So is there no way to add an integer variable to a gradient-based driver?

Justin Gray Over a year ago

By definition, no. A gradient based optimizer requires gradients (derivatives), which are not defined for integer variables.

thearn · Accepted Answer · 2016-02-17 21:43:03Z

1

I believe this error is being raised because the default value for childWeight is provided as the integer 240 instead of a float 240.0. If you ensure that your design variables always have default values as floats, they should not be categorized as a pass_by_obj variable.

answered Feb 17, 2016 at 21:43

thearn

3233 silver badges8 bronze badges

Collectives™ on Stack Overflow

How can I add an integer variable to an openmdao driver?

2 Answers 2

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related