I created a model in Azure ML through AutoML.
Then I selected the "best model":
, then I selected the job that created that model:
, and I finally deployed the model to a real-time endpoint:
With Azure ML v1, the process was a little bit more complicated, and I needed to pass the score.py or generate a new one during the deployment.
Also in this case with v2, however, I need the score.py to customize the Run() function, maybe because I need to extract the predict_proba() rather than just predict() class in a classification model, or for any other reasons.
But I can’t find this score.py in the deployment logs or anywhere else. Where is it? It would also be sufficient a link to a good documentation/article/tutorial that explains how to re-deploy the model with my own scoring file.
2
Answers
I found it: rather than just downloading the registered model, you should rather follow AutomatedML -> Models -> (select your model) -> Download.
This way, the zip contains the three files in the blue rectangle, including the score.py.
If, instead, you just download the model, you get the five files in red, without the score.py.
I'm still not very sure about the reason of this behaviour, however my goal of retrieving the score.py is met.
Regarding the score.py download , you can use the az ml model download – that will get all of the files.