CAPSUL : a pipelining and distributed processing library written in Python

Orateur : Denis RIVIERE (Neurospin, CEA, Saclay)

Capsul (http://brainvisa.info/capsul/) is a free and open-source Python library for encapsulating algorithms and chain them in pipelines. It has unique features for building and dealing with complex pipelines. It is designed to make it easy to execute pipelines in various environments ranging from a local computer to a remote computing center. Capsul is connected to soma-workflow in order to manage all the difficulties of running a lot of pipelines in parallel. Main features : Although written in Python, Capsul does not force algorithms to be written in Python language : pipeline nodes can run any software program. However Capsul provides facilities for Python : a pipeline node can be a simple Python function. A parameters completion system helps to automatically fill parameters from a set of common attributes. An iteration system allows to iterate runs, or pipeline nodes, over a set of parameters configurations. Capsul provides a GUI (graphical user interface) to display pipelines structure in a boxes and links representation, to enter parameters, to iterate runs... The GUI allows basic pipeline edition and graphical building. Pipelines can be saved as XML files. They can thus be written either as Python source files, or as XML files. Pipeline execution can optionally make use of soma-workflow, which manages execution dependencies between pipeline nodes and can run in parallel independent ones, can run and monitor pipelines either locally or remotely on a distant computing resource (a cluster for instance). Capsul has a compatibility with Nipype interfaces which can be directly used as pipeline nodes. A presentation of Anatomist (data visualization) and main pipelines of Brainvisa (http://brainvisa.info/web/index.html) will be done as well.

Télécharger dans mon agenda