Frequently Asked Questions

Which solvers GAMSPy support?

At the moment, GAMSPy supports 21 solvers:

• NLPEC

• SBB

• CONOPT

• CONVERT

• CPLEX

• PATH

• BARON

• CONOPT4

• COPT

• DICOPT

• GUROBI

• HIGHS

• IPOPT

• IPOPTH

• KNITRO

• MINOS

• MOSEK

• SCIP

• SHOT

• SNOPT

• XPRESS

The list can also be accessed from commandline by executing:

gamspy list solvers -a

Or it can be accesed by using the utility function getAvailableSolvers:

import gamspy.utils as utils
print(utils.getAvailableSolvers())

What is the default solver if I don’t specify one?

The listing of default solver for each problem type is below:

Problem	Default Solver
LP	CPLEX
MIP	CPLEX
RMIP	CPLEX
NLP	CONOPT
MCP	PATH
MPEC	NLPEC
CNS	CONOPT
DNLP	CONOPT
RMINLP	CONOPT
MINLP	SBB
QCP	CONOPT
MIQCP	SBB
RMIQCP	CONOPT
EMP	CONVERT

Why can’t I redefine a GAMSPy symbol?

Trying to run the following lines of code will raise an error.

from gamspy import Container, Set, Parameter
m = Container()
p = Set(container=m, name="p", description="products")
price = Parameter(container=m, name="p", domain=p, description="price for product p")

The problem with the above code is that the Set statement creates a symbol in the GAMSPy database with name “p”. Consequently, the namespace “p” is now exclusively reserved for a Set. The following Parameter statement attempts to create a GAMSPy Parameter within the same namespace “p”, which is already reserved for the Set p. Thus, you want to keep in mind that the type for a GAMSPy symbol is fixed once it was declared.

Why do I need a GAMSPy Alias?

Consider the following example code:

from gamspy import Container, Set, Parameter
m = Container()
i = j = Set(container=m, name="i", records=range(3))
p = Parameter(container=m, name="p", domain=[i, j])

p[i, j] = 1

You would probably expect that the value for \(p_{i,j}\) is equal to one for each combination of \((i,j)\)

>>> p.records
            value
    i  j
    0  0   1
    0  1   1
    0  2   1
    1  0   1
    1  1   1
    1  2   1
    2  0   1
    2  1   1
    2  2   1

However, the above lines of code give you:

>>> p.records
              value
    i_0  i_1
    0    0    1
    1    1    1
    2    2    1

Only by declaring j an Alias of i you will get the desired outcome:

from gamspy import Alias, Container, Set, Parameter
m = Container()
i = Set(container=m, name="i", records=range(3))
j = Alias(container=m, name='j', alias_with=i)
p = Parameter(container=m, name="p", domain=[i, j])

p[i, j] = 1

>>> p.records
        value
i  j
0  0   1
1      1
2      1
1  0   1
1      1
2      1
2  0   1
1      1
2      1

Do I use a Parameter or a Python variable to represent scalar parameters?

from gamspy import Container, Parameter, Equation, Sum
m = Container()
p_python = 40
p_parameter = Parameter(container=m, name="p", records=40)

In most of the cases it does not matter whether a scalar Parameter or a Python variable is used. It is more a matter of taste and convenience as:

eq = Equation(container=m, name="eq", domain=i)
eq[i] = Sum(j, x[i, j]) <= p_python

is equivalent to:

eq = Equation(container=m, name="eq", domain=i)
eq[i] = Sum(j, x[i, j]) <= p_parameter

Which functionalities available in GAMS are not (yet) accessible in GAMSPy?

While GAMSPy provides a powerful interface for utilizing GAMS within Python, there are some features from the original GAMS language that are not yet fully accessible in GAMSPy.

Some of the features that have not been fully implemented in GAMSPy include:

1. MPSGE, EMP, EMP-SP:

   Certain specialized GAMS features corresponding to MPSGE, EMP, and EMP-SP are currently not available in GAMSPy. However, efforts are underway to incorporate these features in future updates.

2. Solver-specific features communicated via option files:

   Certain solver-specific features, like indicator constraints, are available in GAMSPy. However, due to the absence of the put facility, generating these constructs may be more challenging. Efforts are being made to provide better ways to communicate such constructs to the solver for enhanced compatibility.

3. External equations:

   GAMSPy does not yet fully support external equations.

4. Extrinsic functions:

   Some extrinsic functions that are part of the GAMS language may not be fully accessible in GAMSPy.

It’s important to emphasize that the GAMSPy team is actively working on expanding the feature set to bridge the gap between GAMS and GAMSPy. If you have specific features or functionalities you would like to see in GAMSPy, feel free to share your feedback with us.

How are GAMS and GAMSPy related?

Dependency

GAMSPy relies on the gamspy_base package, which essentially represents a modularized GAMS installation. When creating a GAMSPy Container, you have the option to specify a GAMS installation independently via the system_directory argument. This enables flexibility in choosing the GAMS version that best suits your needs.

Execution

GAMSPy utilizes the GAMS machinery for critical operations, including the execution of indexed assignment statements, equation definitions, and the solve method. While the typical GAMSPy user does not need to delve into the intricacies of this connection, it’s worth noting that these details may evolve for performance reasons.

Debugging and GAMS Listing File

Although regular Python debugging facilities are usually sufficient, there may be scenarios where additional insights from GAMS prove valuable. If needed, GAMS can provide useful information via the GAMS listing file. For more details on debugging with GAMS, refer to the GAMSPy debugging documentation or the GAMS debugging documentation.

Solver Options

The options for solvers used by GAMSPy are described in the Solver Manuals, which is part of the GAMS Documentation. It’s important to note that examples in the solver manual are based on GAMS syntax, not GAMSPy syntax. When configuring solvers in GAMSPy, users can refer to the relevant sections in the GAMS Documentation for detailed information. as both equation definitions generate \(\sum_{j \in \mathcal{J}} x_{i,j} \le 40\).

However, if you want to change the value of your scalar parameter in between two solve statements like:

from gamspy import Container, Parameter, Equation, Sum
m = Container()
p_python = 40
p_parameter = Parameter(container=m, name="p", records=40)
...
model.solve()
p_python = 50
p_parameter.setRecords(50)
model.solve()

you want to use the GAMSPy Parameter, as changes to a Python variable are not reflected in the generated GAMSPy model. Changes to a GAMSPy symbol, however, will be evaluated by the second solve statement.

Why does pip install gamspy on Google Colab not resolve all dependencies?

Google Colab is a service that allows you to write and execute Python in your browser. It is developed mainly for ML/AI researchers and data scientists. Hence, it comes with many pre-installed packages such as TensorFlow, PyTorch etc. One of these packages is tensorflow-probability==0.22.0 which depends on typing-extensions below version 4.6.0. Since GAMSPy depends on pydantic which requires typing-extensions above 4.6.0, pip does not know how to resolve the issue. This issue has already been resolved by tensorflow-probability in version 0.22.1 (check this). Unfortunately, Google Colab has not update to this version of tensorflow-probability, hence pip complains. This will be fixed as soon as Google Colab updates tensorflow-probability. There are also some packages in Google Colab which don’t have the necessary dependencies in the environment resolved (lida, llmx). To solve all Google Colab related issues run the following:

pip install tensorflow-probability==0.22.1 lida llmx

Now, you can install GAMSPy without any errors:

pip install gamspy