Create MDAs with specific settings

In this example, we show how to create MDAs with specific parameters, such as the maximum number of iterations or the convergence tolerance.

Let us first create two coupled disciplines, namely the Sellar1 and Sellar2 disciplines:

from __future__ import annotations

from gemseo import create_discipline
from gemseo import create_mda

sellar_1, sellar_2 = create_discipline(["Sellar1", "Sellar2"])

Using key/value pairs

A first possibility is to use key/value pairs. In the following code, the tolerance and max_mda_iter settings are passed as key/value pairs. However, this method requires to know the right keyword for the settings, otherwise the create_mda function will raise an error. It is nevertheless relevant for users already aware of the keywords.

mda_jacobi = create_mda(
    "MDAJacobi",
    disciplines=[sellar_1, sellar_2],
    max_mda_iter=15,
    tolerance=1e-8,
)

Using Pydantic settings model

The second way of providing settings for the MDA is to use the corresponding Pydantic settings model. In our example, we intend to create an MDAJacobi, so one needs to import the corresponding model:

from gemseo.settings.mda import MDAJacobi_Settings  # noqa: E402

mda_jacobi_settings = MDAJacobi_Settings(max_mda_iter=15, tolerance=1e-8)

mda_jacobi = create_mda(
    "MDAJacobi",
    disciplines=[sellar_1, sellar_2],
    settings_model=mda_jacobi_settings,
)

Using Pydantic model to provide settings has to be prefered for two main reasons: - The settings are validated (type, value, etc) when the model is created. - The auto-complementation and documentation of the model shows all the settings available for the MDA of interest and their default values.

The name of the Pydantic model associated with an MDA class is available via the class attribute Settings. For instance:

print(mda_jacobi.Settings)

<class 'gemseo.mda.jacobi_settings.MDAJacobi_Settings'>

Updating the settings

It is also possible to update the settings after the MDA object has been created. To do so, one must access the settings model attached to each mda via settings. Then modifications are as simple as:

new_tolerance = 1e-12
mda_jacobi.settings.tolerance = new_tolerance

It is worth knowing that the settings are validated when updated, so the following update of max_mda_iter will raise an error, as the maximum number of iterations must be a non-negative integer.

from pydantic_core import ValidationError  # noqa: E402

try:
    mda_jacobi.settings.max_mda_iter = -2
except ValidationError as error:
    print(error)

1 validation error for MDAJacobi_Settings
max_mda_iter
  Input should be greater than or equal to 0 [type=greater_than_equal, input_value=-2, input_type=int]
    For further information visit https://errors.pydantic.dev/2.9/v/greater_than_equal

Settings for composed MDAs

In GEMSEO, there are two MDAs that are said to be composed because they use or create inner MDAs internally, namely MDAChain and MDASequential. For such MDAs, the settings are nested, since there are settings for these two classes, and settings for the inner-MDAs. Nevertheless, there is a cascading mechanism that allows to update the settings of the inner-MDAs easily. Let us for instance create an MDAChain:

mda_chain = create_mda("MDAChain", disciplines=[sellar_1, sellar_2])

By default, the MDAChain creates MDAJacobi instances internally. To set the tolerance of the inner-MDAs, simply do the following:

inner_mda = mda_chain.inner_mdas[0]
print(f"The tolerance of the inner-MDA is {inner_mda.settings.tolerance}.")
mda_chain.settings.tolerance = 1e-4
print(f"The tolerance of the inner-MDA is now {inner_mda.settings.tolerance}.")

The tolerance of the inner-MDA is 1e-06.
The tolerance of the inner-MDA is now 0.0001.