MS Azure real-time endpoint prediction model usage from Yellowfin

Vefa Gulecyuz shared this idea 6 months ago
Idea Logged

Hello, I have a prediction model deployed in Azure. I have a real-time endpoint to make requests and get back the result of the prediction. I send a get request with the input parameters as a JSON payload. Below is the python format of the get request:

urllib.request.Request(url, body, headers)
Body contains JSON payload as parameters.

I want to be able use this model from Yellowfin.

So far the most suitable way seemed to use JSON connector. However, there have been two problems:

1- JSON connector expects parameters in the url itself. How can pass input parameters in the format JSON payload like above?

2- My prediction model returns either true or false as the response. How can I combine this result with the initial parameters I sent as output?

Or more generally, how can I make use of this model via Yellowfin? I want to send my input from Yellowfin to this endpoint and I want to receive the prediction result back.

Comments (5)

photo
1

Hi Vefa,

Thank you for your patience on this. We are in the process of developing a new JSON Step connector so I will check in on it's ability to pass JSON payloads via the body. I will also do some further testing to see what other options may be available if this is not possible via the new connector. I will come back to you tomorrow with my findings.

Cheers,

Neal

photo
1

Hi Vefa,

Having spoken with the team on the new JSON connector and to clarify it does not currently accept JSON as a REST parameter, rather it uses normal REST and OAUTH to construct the request data in JSON format to be parsed.

However in speaking with one of our consultants it has been suggested to use the R-Plugin instead, which would allow you to script a similar request. A library like this may help you design the required request.

Further information about the R-plugin can be found here on our wiki. If you do require some additional assistance with this please let me know and I will reach out to your CSM to see if we could get a consultant to help further.

Cheers,

Neal

photo
1

Hi Vefa,

Hope you are well. I wanted to touch base and see if there was anything else I could help you with on this?

Cheers,

Neal

photo
1

Hey Neal, thanks for the reply. I have experience with R plugin and I already created a transformation step for this specific case. However, transformation step does not provide the `real time consumption` of a model. I need to schedule the transformation step and it is not that intuitive.

Is there plan to include a feature to YF or to create a plug-in that can allow real time consumption of end points?

photo
1

Hi Vefa,

I apologise for the delay on this one, at this stage I am not aware of this functionality being in the pipeline, however I have created and Idea based off this use case and have submitted it to our developers. Before becoming a request, ideas will be reviewed by our Product Team and chosen for Enhancements based on feasibility, level of positive impact to the user experience, and votes from the community. This post will be updated with any future information relevant to this process.


Cheers,

Neal