¿How to obtain an endpoint from an autoML that gives me the test result in probabilities instead of in binary classes in AzureMLStudio?

Question

I created a model in Azure ML through AutoML.

Then I selected the "best model": image

, then I selected the job that created that model: image , and I finally deployed the model to a real-time endpoint: [image(https://i.stack.imgur.com/vZX1c.png)

But when I test the model it returns to me the binary class of the test but not the probability of belonging to this class. image

I know that for this I have to modify the way it predicts the values, I know that I need to use the predict_proba function.But this automated process of using the best automl model does not allow me to choose the way to obtain the output or at least I do not know it.

I have tried downloading the model once registered and modifying the score.py script and uploading the model again in custom format, but I don't understand how to then upload the environment .yml part. I have also tried to do the entire process only with python from notebooks and I can't.

Does anyone know a simple idea to get the probability in the inferences made on the endpoint instead of a binary class as a result? eternally grateful, even accurate documentation would be welcome, I think I have read it all already

Does anyone know a simple idea to get the probability in the inferences made on the endpoint instead of a binary class as a result? eternally grateful, even accurate documentation would be welcome, I think I have read it all already

Answer 1

I have deployed an endpoint using this sample notebook.

As the score.py contains method in input schema shown below,

You can pass method predict_proba along with your data to get probabilities.