Deploying IBM WKS experimental rule-based models in production: is it advisable to do that?

Question

I would like to know whether it is advisable to deploy in production a rule-based model created with IBM Watson Knowledge Studio (WKS), as it is an experimental feature.

IBM documentation clearly recommends not to use its experimental features in production: https://console.bluemix.net/docs/services/watson-knowledge-studio/troubleshooting.html#experimental .

However, an old post IBM Watson Knowledge Studio 2.0 - deploying a rule-based model is experimental. What does that mean? seems to guarantee that this feature is actually stable and won't be removed in the future. At the same time a more recent post at https://developer.ibm.com/answers/questions/440983/is-my-wks-experimental-data-lost/ shows what happened to someone who deployed its experimental WKS project and then lost it (even though that post is not about rule-based models).

Thank you in advance!

Answer 1

Thank you for your question. I re-read my original answer and I admit that it was a bit ambiguous, for which I apologize. I updated my original answer. To be more clear:

• IBM experimental services and experimental features are NOT suitable for production use. Period.
• We encourage using experimental services/features because the purpose of experimental releases is to learn from client's real use. But this encouragement to experiment with non-GA services/features still does not mean that they are suitable for production use.
• The Knowledge Studio rules-only models, deployable to runtime services like Natural Language Understanding and Discovery, are Experimental. The Rules within Knowledge Studio (rules editor and rules pre-annotator) is a GA feature. This is why I was saying that Rules are here to stay, while the rules-only models are not protected agains breaking changes. For example, we may decide to pull off the market the rules-only models when we introduce hybrid models (rules + machine-learning in the same runtime model) or if we don't see good adoption of the rules-only models. At the same time the Rules-based pre-annotation has proven to be a valuable and well accepted feature that we plan to enhance.

My apologies for the initial confusion. I hope that this answer is more clear than my previous one.

Kind regards,

Stefan

Answer 2

Experimental features means a number of things in relation to a production solution.

• How it works may change at a later time.
• No guarantee it will be available at a later time.
• It is generally unsupported if issues occur.

One of your links refers to "Experimental service". These should never be used in a production environment, because when they go live the experimental service will stop working. Live versions may not be fully compatible as well.