Azure Machine Learning
An Azure machine learning service for building and deploying models.
3,332 questions
@Ramr-msft @romungi-MSFT Is there somthing new here? Or do I have to create a suppor ticket?
Use register trained .ilearner model and deploy it as Real-Time-Inference Endpoint
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(153.6, 91%, 24%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EMS%3C/text%3E%3C/svg%3E)
Michael Strohmayer
96
Reputation points
Dear Community, since more than 2 weeks I'm struggeling on Azure Machine Learning Studio. Our Training-Pipeline generates a Trained-Best-Model folder containing the following files _meta.yaml _samples.json _schema.json conda_env.yaml data.ilearner model_spec.yaml score.py In the designer I can just run my training Pipeline, Update my Inference Pipeline and´once this finishied the progress, I press the "Deploy" Button. This creates some kind of deployment package, registers the model and deploys it to a kubernetes cluster. I'm completly happy with the setting of my pipeline and my resulting endpoint. But our customer wants us to automate it so there is a weekly deployment shedule on the endpoint. All tutorial and informations I found use other file formats (like .pkl and .onnx) but really not a single Jupyter notebook shows me how a) To Read my mltable Dataset (Type File, contains the folder of the listed filnames above ) and "convert it" to a model b) package this model c) deploy it to an existing kubernetes webservice If i could just automate those 2 clicks from the Inference Pipeline Run in python, it would all be done. But this issue already consumed more like 2 weeks of constant failing. Is it so hard or is it just me? ![205506-image.png][1] [1]: /api/attachments/205506-image.png?platform=QnA
Azure Machine Learning
{count} votes
-
romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator2022-05-26T08:00:52.23+00:00 anonymous user To automate the pipeline runs I would recommend to review the option to use Azure DevOps pipelines with Azure ML Pipelines. The sample setup to achieve this takes less than an hour to setup, you can follow steps in this repo to get it up and running.
Since you are using the studio, the train model component produces the .iLearner binary format which I think cannot be modified directly. Please see here from documentation.
What training produces is a specific binary format, the iLearner, that encapsulates the statistical patterns learned from the data. You cannot directly modify or read this format; however, other components can use this trained model.
-
Michael Strohmayer 96 Reputation points
2022-05-26T15:50:09.62+00:00 Thanks for your answer, but I can't really understand why I can deploy the ilearner model to an endpoint using the designer and can not achive this with python. If i download a model that has been registered with the designer, rename it as .zip and look at the components, I see, that it contains exactly the files that i'm getting due to my training. So I guess, the designer is actually just calling a package-command and a deploy command. How wouldI use the package command https://learn.microsoft.com/en-us/python/api/azureml-core/azureml.core.model.model?view=azure-ml-py#azureml-core-model-model-package on my files listed above to be able to register the model? Once I have the Model registered, how am I able to deploy it as an inference Endpoint to a kubernets cluster (in the same way as the Designer does it?) [1]: /api/attachments/205872-image.png?platform=QnA [2]: /api/attachments/205869-2022-05-26-13-00-24.png?platform=QnA
-
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(89.60000000000001, 7.000000000000001%, 18%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EA%3C/text%3E%3C/svg%3E)
Alexandre 46 Reputation points2022-05-31T09:17:07.893+00:00 Hi Rohit and Micheal
Is ilearn a way to copy the model? I am looking for ways to migrate model from old studio to the one in azure portal. Any idea?
-
Michael Strohmayer 96 Reputation points
2022-06-02T06:11:33.32+00:00 Those ".ilearner" files are used by the Azure ML Designer and they contain all the relevant data of your model. But it seems that in the whole machine learning world only microsoft uses those. As you can read in this question, not even the Microsoft Support Team know how to work with those models. If you can, stay away from those models and dont really use the Designer models at all...
-
Michael Strohmayer 96 Reputation points
2022-06-02T06:12:58.75+00:00 @romungi-MSFT do you have any further information on this here? I just want to register my ilearner files as a model, package it and deploy it with python. Is there really no way to achive this?
Sign in to comment