Call external solver during SP solve

[bqpd]

These should be tested, written, used:

For changing a constant:

import numpy as np
from gpkit.constraint.set import ConstraintSet

class ChangeConstantConstraint(ConstraintSet):

    def as_gpconstr(self, x0):
        gpconstr = super(ConstraintSet, self).as_gpconstr(x0)
        if x0["Re"] >= 1e6:
            gpconstr.substitution.update(xfoil(x0))
        return gpconstr

def xfoil(x0):
    return {"CD": np.sin(x0["Re"])}

For inserting a constraint:

def xfoil(x0):
    return [PosynomialInequality, MonomialEquality]

class ChangeConstantConstraint(ConstraintSet):

    def as_gpconstr(self, x0):
        gpconstr = super(ConstraintSet, self).as_gpconstr(x0)
        if x0["Re"] >= 1e6:
            constraints = xfoil(x0)
        else:
            constraints = [PosynomialInequality, MonomialEquality]
        return ConstraintSet([gpconstr, constraints])

[bqpd]

Alternative conditions for calling the outside function, besides variable bounds include

• sensitivity bounds (upper and lower)

[codykarcher]

What file should these things live in?

[bqpd]

These should live in one of your code files, where they can reference variables and so on. You can wrap all your other constraints in one of these (for the substitutions one) or just the relevant constraints (for the "add a bonus constraint" one).

[codykarcher]

I just pulled, but I have no module named constraint? Not sure what's going on here. I also can't find ConstraintSet in the source code...

[codykarcher]

Here's a MWE we can reference:

from gpkit import VectorVariable, Variable, Model, units
import gpkit
from gpkit.constraint import ConstraintSet

Re = Variable("Re", 1e7, "-", "Reynolds Number")
CL = Variable("C_L", 1.0, "-", "Lift Coefficient")
CD = Variable("C_D", "-", "Drag Coefficient")

def xfoil(x0):
    return [CD >= 0.01 + .03*CL**2]

class AeroModel(ConstraintSet):

    def as_gpconstr(self, x0):
        gpconstr = super(ConstraintSet, self).as_gpconstr(x0)
        if x0["Re"] >= 1e6:
            constraints = xfoil(x0)
        else:
            constraints = [CD >= 1. + CL**2]
        return ConstraintSet([gpconstr, constraints])

constraints = AeroModel.as_gpconstr([])

objective = CD

m = Model(objective, constraints)
m.solve('cvxopt', verbosity=1)

Should return CD=0.04 in its current state and CD=2.0 when we drop the Re beneath the cutoff. Pretty convinced this is not how it was intended to be implemented, but it's my best guess.

[bqpd]

Great example. Updated and working on my computer on the "externalsolver" branch:

from gpkit import VectorVariable, Variable, units, SignomialProgram
import gpkit
from gpkit.constraints.set import ConstraintSet

Re = Variable("Re", "-", "Reynolds Number")
CL = Variable("C_L", "-", "Lift Coefficient")
CD = Variable("C_D", "-", "Drag Coefficient")

def xfoil(x0):
    return [CD >= 0.01 + .03*CL**2]

class AeroModel(ConstraintSet):

    def as_gpconstr(self, x0):
        gpconstr = super(AeroModel, self).as_gpconstr(x0)
        if x0 and x0["Re"] >= 1e6:
            constraints = xfoil(x0)
        else:
            constraints = [CD >= 1. + CL**2]
        return ConstraintSet([gpconstr, constraints])

constraints = AeroModel([CL >= 1.0, Re == 1e7])

objective = CD

m = SignomialProgram(objective, constraints, require_signomial=False)
m.localsolve('cvxopt', verbosity=1)

[codykarcher]

Works for me now too. I'm going to close and raise new issues as they come up. Thanks @bqpd

[codykarcher]

nvm, I can't close the issue... @whoburg

[whoburg]

Looking great. Exciting that we can do this so cleanly.

Remaining todos:

• [x] add example of this to readthedocs

[whoburg]

shall we close this now that @cjk12 implemented #687 ?

[bqpd]

Looks great! http://gpkit.readthedocs.org/en/latest/signomialprogramming.html#calling-to-external-codes