5

I recently built a classifier using the built-in Classify command in MMA v11.01. We are exploring the feasibility of using python-based workflow (to be consistent with our company practice) that would allow us to use this model (a neural net) to score a set of customers each week, using data drawn from our internal SQL-based data warehouse, and then pass the scored results back into our primary sales system. Ideally we would simply invoke the classifier via a python script, which would provide the input data and a link back to the final destination.

In doing a bit of research about this, there are some examples of integrating python and MMA in some fashion. But many of the examples seem very dated, or show very trivial implementations (e.g., invoke a single MMA function). In addition most of the examples I've found suggest the python-MMA "link" is not really supported and may not be reliable for this sort of application.

I'm curious is anyone has developed this sort of link using python, particularly for a production corporate application. I suppose that this could also be done via the Wolfram Cloud, though that raises some other technical issues for my company. As WRI seems to be pushing hard into datas science, where python is one of the predominant languages, I'm hoping there is an "official" -- or at least a WRI sanctioned -- solution.

Thanks,

Mark

MSC02476
  • 699
  • 4
  • 12
  • Which operating system are you using? It is easy to run Python scripts from within Mathematica (using RunProcess[$SystemShell,All, "mypythonscript"], even on Windows. Probably it is also not hard to use such a simple connection by using WolframScript from within Python. – Rolf Mertig Oct 12 '16 at 19:47
  • Hi @RolfMertig. Thank you for your comment. This specific application would be in a Windows environment. We have others applications at my company would be based in Linux. Like many researchers in a corporate environment, I'm restricted to what tools can be used in production. So MMA cannot be part of our production workflow, even though we use it extensively for our analytical work. Thus my question is about feasibility of using MMA output, rather than invoking MMA itself. I hope this helps. – MSC02476 Oct 14 '16 at 14:50
  • You could just write the Mathematica results (can you give a specific example?) to hard disk or better directly to the database from within Mathematica and then go on with Python. – Rolf Mertig Oct 14 '16 at 20:30

0 Answers0