Solver Extension

GEKKO includes a limited interface to access more solvers. The solver extension module allows for converting the GEKKO model to other mathematical optimization libraries, opening up access to more solvers. It currently contains two converters, allowing the GEKKO model to be solved through AMPLPY or Pyomo.

Setup

To use the solver extension module:

  • Set the SOLVER_EXTENSION option to the converter you want to use (either AMPLPY or PYOMO)

    • m.options.SOLVER_EXTENSION = <converter>

  • Set the SOLVER option to the solver you want to use (eg. ipopt). Note that the string specifying the solver may be case sensitive.

    • m.options.SOLVER = <solver>

  • The GEKKO model can be declared like normal. The results from the solve are placed back into the GEKKO model variables.

Solver Options

Solver options are specified within m.solver_options as normal:

# Use options relevant to the solver you are using.
m.solver_options = ['max_iter 10', \
                    'tol 0.01', \
                    'outlev 1' \
                    # etc...
                    ]

AMPLPY

The solver extension module supports converting to AMPL syntax, allowing access to various supported solvers by making use of the AMPLPY library. Alternatively, a .mod file (AMPL model file) can be output and solved by uploading to NEOS.

The base version of AMPL limits a model to 500 variables and 500 constraints (300 for nonlinear problems, and fewer for certain solvers). AMPL offers a free Community Edition license with no limitations on variables or constraints. Licensing and more details can be obtained from the AMPL website.

The converter between the GEKKO model and AMPLPY is limited and does not support the full range of model building functions and options available in GEKKO. However, basic model building functions such as (but not limited to) variables, parameters, constants, intermediates, constraints, and objectives are supported by the converter. Functions relating to dynamic optimization (time, derivatives, etc) are not supported by AMPLPY and cannot be used in the converter.

Setup

The solver extension module requires AMPLPY to solve within GEKKO:

$ pip install amplpy

Solvers are installed through amplpy.modules. See https://dev.ampl.com/ampl/python/modules.html:

$ python -m amplpy.modules install <solver>

AMPLPY Example

Example use of the solver extension module with the AMPLPY converter is shown below:

from gekko import GEKKO
m = GEKKO(remote=False)  # remote=True not supported
x = m.Var()
y = m.Var()
m.Equations([3*x+2*y==1, x+2*y==0])
# enable solver extension and use AMPLPY converter
m.options.SOLVER_EXTENSION = "AMPLPY"
m.options.SOLVER = "bonmin"  # use BONMIN solver
m.solve()    # solve
print(x.value,y.value)

Additional AMPLPY converter methods

The AMPLPY converter provides some utility methods to the GEKKO model object:

classmethod m.create_amplpy_object()

Returns an amplpy model object:

ampl = m.create_amplpy_object()
# do some stuff with the amplpy object
#...

For more information view the amplpy documentation.

classmethod m.generate_ampl_file(filename='model.mod')

Generates an ampl model (.mod) file in the current directory or as specified by filename:

m.generate_ampl_file()

Pyomo

The solver extension module supports converting to Pyomo, a python library for mathematical optimization. Pyomo supports a variety of solvers, including an interface with ASL (AMPL Solver Library). One of the primary advantages of Pyomo is that it is entirely open source (BSD license), and therefore has no constraints on model size and is free for commercial use.

The converter to Pyomo supports the basic GEKKO model building functions such as variables, parameters, constants, intermediates, constraints, and objectives. In its current state, it does not support dynamic optimization (time, derivatives, etc). However, this has the potential to be implemented in the future through use of Pyomo DAE (Differential Algebraic Equations) module.

Setup

The solver extension module requires Pyomo to solve within GEKKO:

$ pip install Pyomo

Solvers should be installed by either compiling them from source or obtaining the relevant binaries. The location of the executable should then be added to PATH in order for Pyomo to recognise the solver.

You can check if the solver has been installed correctly by running a version check, ie. <solver> -v.

Pyomo Example

Example use of the solver extension module with the Pyomo converter is shown below:

from gekko import GEKKO
m = GEKKO(remote=False)  # remote=True not supported
x = m.Var()
y = m.Var()
m.Equations([3*x+2*y==1, x+2*y==0])
# enable solver extension and use Pyomo converter
m.options.SOLVER_EXTENSION = "PYOMO"
m.options.SOLVER = "cbc"  # use CBC solver
m.solve()    # solve
print(x.value,y.value)

Additional Pyomo converter methods

The Pyomo converter provides some utility methods to the GEKKO model object:

classmethod m.create_pyomo_object()

Returns an pyomo ConcreteModel object:

pyomo_model = m.create_pyomo_object()
# do some stuff with the pyomo ConcreteModel
#...

For more information view the Pyomo documentation.