In CP4D analytics project, there is no model release management framework like SPSS Collaboration and Deployment Services (CaDS).
1. LEARNING - Running batch script for model re-training at some intervals(weekly or monthly), and save new model as the LATEST version.
2. EVALUATION - Running batch script for model evaluation that executes scoring model (both LATEST and PRODUCTION version) with test data and save predicted data to database. Analyst can check the result of predicted data (MAE, MAPE, and predicted data graphs) on BI tool after running batch.
3. RELEASING NEW VERSION - Running batch script of replacing PRODUCTION version. This batch makes LATEST version to PRODUCTION.
In case of batch scoring, there must be unique named scoring model/script because this scoring script is called by other program out of CP4D.
In case of real time scoring (web-service scoring), there must be unique URL for model/script because this URL is called by other program out of CP4D.
CaDS can address to these requirements by using CadS label feature. I use "LATEST" label to latest version, and "PRODUCTION" label to production version. After LEARNING, the LATEST label moves to the new version, but PRODUCTION label is kept on the version without moving. I can run RELEASING NEW VERSION batch script to move PRODUCTION label to any version I specified.
CP4D seems not to have these feature. Every creating new version of model, I have to create new project release with another name. It causes that the scoring web service URL of new version is not same as old version, so I have to change our scoring program that send REST request to the scoring URL in CP4D.
I strongly hope CP4D to add model release version management feature such as CaDS.
|Who would benefit from this IDEA?||All person who use CP4D for continuous model developing and deployment.|
|Priority Justification||It is necessary for continuous model creation and deployment system such as CP4D.|
How should it work?
(added on 9th Mar 2020)
I think these 4 detailed features are needed for above requirements.
(1) Same endpoint URL for both old and new version of online deployments
(2) Deployment versioning in Deployment Space
(3) Model versioning in Deployment Space
(4) Any version of deployments can be switched as "in use" at any time.
For the details of (4), please refer the sample scenarios in attached file.
|IBM's success depends on gathering feedback from customers like yourself. Aha Ideas Portal is the third party tool through which IBM Offering Managers gather feedback from customers such as yourself.|
|IBM is a global organization with business processes, management structures, technical systems and service provider networks that cross borders. As such, the information collected through Aha Ideas Portal (Customer Name, Customer Email Address) will be stored by them in the United States, and handled only as per IBM's instructions and policies. Your data (Name and Email Address) will NOT be shared with other IBM customers.|
|In order to safeguard your information in Aha, do not leave your workstation unattended while using this application, log off after using it, and print only if necessary. If you need to make a hardcopy, remember to pick up the print-out immediately, keep it under lock, and destroy it immediately when no longer needed.|
|NOTICE TO EU RESIDENTS: per EU Data Protection Policy, if you wish to remove your personal information from the IBM ideas portal, please login to the ideas portal using your previously registered information then change your email to "firstname.lastname@example.org" and first name to "anonymous" and last name to "anonymous". This will ensure that IBM will not send any emails to you about all idea submissions|