问题
When building a rule-based model in Watson Knowledge Studio, a reminder appears saying:-
The rule-based model that you build here can be deployed and used by other Watson services for experimental purposes only.
In the online documentation, at the beginning of chapter "Deploying a rule-based annotator to IBM Watson Discovery", another similar mention appears:-
Attention: This is currently an experimental feature of the service.
I started building an application based on a WKS custom rule-based model deployed in Discovery and i want to be sure this solution will not be jeopardized at some point in some future release of either WKS or Discovery.
What experimental means in this context? Is there a chance that in a future version of WKS or Discovery the rule-based modeling won't be available anymore?
回答1:
I'm the Offering Manager for WKS.
An example of the official definition of Experimental/Beta for IBM can be found here: http://www.ibm.com/software/sla/sladb.nsf/pdf/6605-11/$file/i126-6605-11_06-2017_en_US.pdf
My interpretation of the official IBM policy is as follows:
- Experimental/Beta provides no warranty
- Experimental/Beta does not guarantee performance and is not suitable for production
- Experimental/Beta does not provide migration to GA
- Experimental/Beta does not save client data at service/feature termination - client has to save whatever data they want to retain
- Experimental/Beta can terminate without prior notice
- Any feedback from Experimental/Beta client can be used by IBM and third parties
In the context of WKS, we decided to keep the Rules-only models as an Experimental feature for the time being because we are still experimenting around how to combine the rules models with the machine-learning models. (Currently, they are separate, which decreases the value for the WKS users.) The Rule-based technology is stable as a whole and I can assure you that it is not going away. But we want to improve the UX around it. We'll also make our best effort to protect any client investments in rule-based model, although we reserve the right to introduce breaking changes if we have to. (Hence, the "Experimental" state of the feature.)
There are two aspects of using Rules in WKS:
- As a pre-annotator for the machine-learning model annotation
- As stand-alone (deployable to Natural Language Understanding and to Watson Discovery services) rule-based model.
Because rules-only models are still in Experimental state, you benefit from using them in production at no charge even when you deploy the rules-only models to paid IBM Cloud services.
I would encourage you to continue using the Rules in WKS and to provide us with feedback on ways to improve the feature. As I mentioned, the rules are here to stay and we are working on improving them.
Thanks for your question and for using WKS.
Kind regards,
Stefan
来源：https://stackoverflow.com/questions/45302101/ibm-watson-knowledge-studio-2-0-deploying-a-rule-based-model-is-experimental