Namespaces¶

What are namespaces?¶

Namespaces are prefixes to input or output names of the MDODiscipline subclasses. The name of the variable is replaced by the namespace, a separator, ‘:’ by default, and the original variable name.

This allows to control the data exchanges between existing disciplines, and therefore configure the process without changing the original grammar. For instance, if a discipline A with input “x” and output “y” is chained with a discipline B of input “y” and output “z”, adding the namespace “ns” to the output y of A will result in a “disconnection” between “y” as the output of A and “y” as the input of B.

In terms of API this would result in the following example:

>>> from gemseo.api import create_discipline
>>> from gemseo.core.chain import MDOChain
>>>
>>> def func_a(x=1.):
>>>     y=x+1
>>>     return y
>>>
>>> def func_b(y=1.):
>>>     z=y+1
>>>     return z
>>>
>>> a_disc = create_discipline("AutoPyDiscipline", py_func=func_a)
>>> b_disc = create_discipline("AutoPyDiscipline", py_func=func_b)
>>> print("A inputs", list(a_disc.get_input_data_names()), ", outputs",
...    list(a_disc.get_output_data_names()))
A inputs ['x'], outputs ['y']

>>> chain = MDOChain(disciplines=[a_disc, b_disc])
>>> print("Chain output, z = ", chain.execute()['z'])
Chain output, z = [3.]

>>> a_disc_ns = create_discipline("AutoPyDiscipline", py_func=func_a)
>>> a_disc_ns.add_namespace_to_output("y", "ns")
>>>
>>> print("A inputs", list(a_disc_ns.get_input_data_names()), ", outputs",
...     list(a_disc_ns.get_output_data_names()))
A inputs ['x'], outputs ['ns:y']

>>> chain_ns = MDOChain(disciplines=[a_disc_ns, b_disc])
>>> print("Chain output with namespace, z = ", chain_ns.execute()['z'])
Chain output with namespace, z = [2.]

Warning

This is an experimental feature, that is currently validated for the main process classes: MDOChain, MDA and its subclasses, MDOParallelChain etc. Scenarios can be created with disciplines handling namespaces. The main limitation is that not all wrappers and MDO test problems are compatible with namespaces, which requires the modifications described at the end of this page. Currently, the AutoPyDiscipline and ConstrAggegationDisc support namespaces and can be used as examples.

Coupling control in MDAs¶

Namespaces allow to control the couplings in MDAs by renaming a variable. This may change the coupling structure graph, as illustrated in the next figure.

../_images/namespaces_and_coupling.png — Controlling the couplings using namespaces.¶

Impact on the MDODiscipline wrappers¶

The discipline that wraps a simulation code, such as AutoPyDiscipline, declares its input and output names without the namespace prefix, in MDODiscipline.__init__().

After instantiation, a namespace may be added to the discipline, which may make the names of the grammar elements inconsistent with the names of the local variables in the discipline wrapper. The wrappers must be adapted to handle input names changes due to namespaces. To this aim, the DisciplineData values may be accessed from keys with or without namespaces. Also MDODiscipline.get_input_data_names() has an argument “input_prefix” that allows to define whether to return namespace prefixes or not.

Besides, BaseGrammar has the attributes BaseGrammar.to_namespaced and BaseGrammar.from_namespaced that map the names with and without namespace prefixes.

Finally, MDODiscipline.store_local_data() allows to pass variables names without namespace prefixes. This allows to adapt wrappers to support namespaces with only minor modifications.

For instance, the AutoPyDiscipline._run() method is as follows, and supports namespaces:

def _run(self):
    output_values = self.py_func(**self.get_input_data(namespaces_prefix=False))
    self.store_local_data(**output_values)