You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One of the goals of AgriFoodPy is to provide an accessible tool to model agrifood systems.
A pipeline manager can help us achieve this by closing the gap between the non-expert user wanting to test the effect of certain interventions to the food system, and the low level API we are developing.
My very basic view of the pipeline manager is as follows:
The user generates a configuration file describing the list of interventions, models, outputs
Each of these is described by a set of parameters. These can be set side by side with the above list
Interventions, models, chosen outputs, etc, are applied/extracted from a set of provided data.
A known example of how this could work is CosmoSIS, which is used in Astrophysics as a parameter inference tool. It connects a sampler with a series of user selected modules which run in sequence and return a likelihood to the sampler.
The part I would be interested in replicating is the module sequence which, in very simple terms, works by reading and writing data to a central structure called the Datablock.
In AgriFoodPy, our Datablock can be a dictionary containing the different datasets each module will require, the metrics and outputs of each module / model. These outputs can be used as intermediate input data for the next modules in the pipeline, or can be extracted at the end to be analysed, plotted, etc.
I'm not aware of any generic tool to facilitate this. As far as I know, CosmoSIS was written from scratch by Joe Zuntz.
If it follows a well known strategy or paradigm, we could explore that instead.
Ian suggested some pipeline management tools by email some time ago:
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
One of the goals of AgriFoodPy is to provide an accessible tool to model agrifood systems.
A pipeline manager can help us achieve this by closing the gap between the non-expert user wanting to test the effect of certain interventions to the food system, and the low level API we are developing.
My very basic view of the pipeline manager is as follows:
A known example of how this could work is CosmoSIS, which is used in Astrophysics as a parameter inference tool. It connects a sampler with a series of user selected modules which run in sequence and return a likelihood to the sampler.
The part I would be interested in replicating is the module sequence which, in very simple terms, works by reading and writing data to a central structure called the Datablock.
In AgriFoodPy, our Datablock can be a dictionary containing the different datasets each module will require, the metrics and outputs of each module / model. These outputs can be used as intermediate input data for the next modules in the pipeline, or can be extracted at the end to be analysed, plotted, etc.
I'm not aware of any generic tool to facilitate this. As far as I know, CosmoSIS was written from scratch by Joe Zuntz.
If it follows a well known strategy or paradigm, we could explore that instead.
Ian suggested some pipeline management tools by email some time ago:
I will start investigating these in detail now. We can use the Dashboard as a testing canvas for pipeline execution.
JP
Beta Was this translation helpful? Give feedback.
All reactions