Pyphant – A Python Framework for Modelling Reusable Information Processing Tasks

Abstract

We are presenting the Python framework “Pyphant” for the creation and application of information flow models. The central idea of this approach is to encapsulate each data processing step in one unit which we call a worker. A worker receives input via sockets and provides the results of its data processing via plugs. These can be connected to other workers' sockets. The resulting directed graph is called a recipe. Classes for these objects comprise the Pyphant core. To implement the actual processing steps, Pyphant relies on third-party plug-ins which extend the basic worker class and can be distributed as Python eggs. On top of the core, Pyphant offers an information exchange layer which facilitates the interoperability of the workers, using Numpy objects. A third layer comprises textual and graphical user interfaces. The former allows for the batch processing of data and the latter allows for the interactive construction of recipes.

This paper discusses the Pyphant framework and presents an example recipe for determining the length scale of aggregated polymeric phases, building an amphiphilic conetwork from an Atomic Force Microscopy (AFM) phase mode image.