Provide model viewers in ML Studio
In the SQL Server data mining we can inspect the models, whereas in AzureML we can only evaluate and score models: they have become black boxes. In many industries it is important to not only predict values, but also have an explanation on why these predictions where made.
I understand that visualizations for decision jungles might be more complex than the simpler decision trees we have in SQL Server, but to convince people in using machine learning, we need to go beyond black box models.
Hi Azure Customer,
We will close this for now since it has been 4 years and the product changes a lot. We are actively moderate feedback in User Voice. Please keeps giving us your great feedback, we always considerate it seriously ^^
Thank you for understanding.
Azure CXP Community
Spoke with Parmita Mehta after her AzureML session at TechEd Europe and she said that some sort of model explanation was "being considered". Unfortunately nothing yet in the GA version.
Phil Z commented
Strongly support. Need to be able to share underlying intuition of models to build support for using ML. I also see this as great for teaching ML and model viewing is key to teaching underlying intuition. Would also love traditional regression diagnostic reports for similar reasons.
Rui Quintino commented
Agree, I understand that making it a white-box model can raise some issues on the business model for azure ml... but not completely comfortable with the black box model, inhibits some very interesting scenarios where you may not be immediately looking for a score model but to accelerate data discovery/model evaluation/inspection, a roadmap would be nice, will it support in the future?