AutoML with Auto-Sklearn ❌ Automated Machine Learning with Auto-Sklearn

[@jacksonrudd3886]

Why does ensembling result in overfitting?

[@richarda1630]

I agree with you that I think the initial "dirty work" of data gathering , cleaning and preparation will not be automated anytime soon (but again this is scenario/industry specific) But the currently higher-order of skills which are now the purview of the data scientist will be automated. Given that scenario, what advice and or learning strategy would you give to budding data scientists who are just getting into the field? Should they go straight to AutoML? Awaiting your insights , thanks !

[@DecisionForest]

Good point, AutoML requires a good foundation regardless whether the optimization is done automatically. Because that foundation leads you to finding the right solution to the problem. So I'd say as you learn a new topic, try to find practical problems that you can solve with it. That will help train your problem solving abilities which many new data scientists lack.

[@zaferemrekilinc4138]

a very informative and impressive video

[@DecisionForest]

Thank you 🙏

[@dr.cancan]

Hi thank you so much great video! I have one question, Can we interpret auto-sklearn and the contribution of predictor features/columns for the selected model?

[@DecisionForest]

I personally haven't as I'm not using auto ML on a regular, but I think you should be able to use LIME with the classifier.

[@Infinity-ho9qv]

Thanks, it was helpful...

[@ramons.g5135]

Great video! I think auto-sklearn is nice to have, but I do not think it is going to replace us. I can see that it automatically refits with the best model, but what if you want to choose a slightly underperforming one, but that adds regularisation and hence is less prone to overfitting on unseen data, or a more interpretable one so you can do inference as well? I think it will help us decide which one we should go for, but the understanding of the pros/cons of each model ultimately relies on the data scientist, and critical thinking will never be automated :)

[@DecisionForest]

You're spot on! The case you described is perfectly representative of what many don't realise about our job. That creativity and critical thinking are crucial to success in this field.

[@ramons.g5135]

Exactly, treating ML as a black box that churns out numbers is a huge mistake. By the way, I am looking for a job, if you know anyone hiring please let me know :)

[@DecisionForest]

@@ramons.g5135 I'd love to help but don't have much crossover with hiring managers. Your best chance is to connect to as many data science recruiters from your area on Linkedin and take it from there.

[@aryandavoudi3477]

It's really amazing!! BUT can we do cross validation with auto-sklearn?

[@DecisionForest]

You should find some examples here: automl.github.io/auto-sklearn/master/examples/