05-23-2018 12:08 PM - edited 08-03-2021 11:24 AM
Want to get started with Promote, but don't know Python?
This article describes:
Quick question, and I'm not sure where to put it. For operationalizing a model, one thing we're seeing is that the business area just has their business data. A modeling team will mix in various additional data (e.g. say, demographics), in order to build a better model... they may also do PCA and various other feature engineering and reduction). So, when it comes time to score, the business system does not possess all the engineered columns required by the model. We need to do, what we're referring to as "external data joins" at scoring time.
The easy solution for this is to tell the business system too bad, so sad: you need to jump through hoops to build the correct columns.
Long story short. We don't have Alteryx Server, but my understanding is that if we did, we could deploy a full-on workflow that takes in the business area data, does the "external data joins" and column development... and then, heck since we're already in a full-on workflow, why not just include the scoring process there and return the score, perhaps with additional useful info (error handling, score interpretation, etc)? [And this is all available via REST API on Alteryx Server.]
Is that a reasonable thought process? Thanks!
Yes, you are on the right path.
There are a few options:
1. You could build a full workflow that does all the prep work and joins.
2. You could score the model with the Python SDK tool, or if the model is deployed to Promote, the prepared data could be sent to Promote for scoring. You could also build all the external joins using the Python SDK tool.
With Alteryx Server, the entire workflow could be called using the API. Alternatively, you could build an analytic app where a user could add their data and run the workflow on the Server (without using the API).