Skip to content

Filter Pipelines and Filter Sequences

Felix Maurer edited this page Jul 11, 2021 · 3 revisions

Filter Pipelines

The filtering system contains two Filter Pipelines, applied at two different points in time during the coupling procedure.

  • The first of the two is applied per MD instance just before that instance's data gets averaged to then be send to the macro solver. This pipeline is referred to as per-instance.
  • The other one is used slightly afterwards, after the MD instances' data has already gotten averaged. It is thus called post multi-instance.
    From a user's perspective, the two behave identical. For information on how to configure a pipeline, cf. Configuration: Filtering.

Filter Sequences

A filter pipeline consists of one or more sequences. Such sequences contain one or more filters. Note that while usually a filter takes one set of input MD-To-Macro data and produces and output of exactly the same size, exceptions to this exist in the forms of Filter Junctions and Asymmetrical Filter Junctions. For such exceptions, special kinds of filters called Junctors are used.
This means that for all purposes, Junctors form a subset of filters, and junctions form a subset of sequences. Their complement regarding the entirety of filters/sequences is sometimes referred to as regular or normal filters/sequences.
Below you can find an image showing an exemplary filter pipeline configuration, taken from [1].
TODO: Insert image

Future Plans

We recognize that the term "Filter Pipeline" is not very descriptive and should be refactored to something more fitting.

References

[1] Jarmatz, Piet, Felix Maurer, and Philipp Neumann. "MaMiCo: Non-Local Means Filtering with Flexible Data-Flow for Coupling MD and CFD." International Conference on Computational Science. Springer, Cham, 2021.