Learning Azure Machine Learning Studio

[Ritchie Poon]

Hi i recently completed a predicative experiment for the above and deploy to web service, first on class web, then later on new web.

After input the SAME inputs, I got two results, 149.. (correct based on the DOC guide) vs 59.. (incorrect). I COPIED the experiment from free tier to the other work space to generate the NEW Web Service。

I tried to compared everything and could not find any differences between the TWO Predicative Exp that I used to generate the Web Service (except they are final run on two different dates.)

Finally, I re-deploy the web service once again, and it is now correct. I could not figure out what caused this (just run it again and redeploy should not cause any problem at all).

I could understand in other "non-numerical' problems i may have to redeploy the solution or configuration, but when it comes to such ML thing, there would be no clue leading to a re-deployment, and one must just adopt a "wrong figure".

I can't see what i could have done wrong:

a) Complete Web Service (Classic) in One Work Space

b) Copy the Predicative Exp to Another Work Space

c) Re-run the Predicative Exp

d) Deploy to New Web Service (from New Work Space)

e) Error: I compared and find different results from Classic vs New (Test Response Request)

   I tried both CSV and on-screen input and the SAME ERROR came out of 59 calorie

f) I re-Deploy the New Web Service in New Work Space

g) Then I got the SAME results..

In ML modelling, I dont expect such re-redeployment is the Right Step (as you can't see any bugs...) to solve the problems.

So any potential bugs within the system? I could not see any chance i would miss a step (or if really so, that would still be a Design Flaw that I could render a new deployment that looked "correct" initially.

Thanks

Ritchie Poon