This is how to take your ML models from great to GOAT

Рет қаралды 6,818

ritvikmath

Күн бұрын

Пікірлер: 15

@Darkev77 3 жыл бұрын

Would love to see a video on Bayesian Search and the math behind it. Mad respect for that cap

@halo99yo 3 жыл бұрын

Hyperparameter tuning is an integral part of using ML. The reason it "doesn't get a lot of love" is because of all the people coming from programming or business who's never taken a proper university course in statistics or ML. Don't be surprised if they have no idea that these hyperparameters even existed. But like another comment mentioned, the corporate world is just hiring any programmer that knows python as a "data scientist" even if they don't have the necessary background in Statistics.

@junkbingo4482 3 жыл бұрын

right man, we definitely think the same ( and i'm lots older than you and taught at the university...) sad world, with HR twats hiring people with skills, but not necessarily the good ones a class in python never fixed a pb but thats a thing you cannot explain to people who dont want to get it cheers

@chrislin5938 2 жыл бұрын

This idea really blew my mind. Thank you so much for creating these videos!! Respect

@amitagni2 3 жыл бұрын

Thanks for the video. I've got a question. In your experience, how many iterations would you usually set for bayesian search?

@deltamico Жыл бұрын

Plus the shape of the hyper parameter results table is given by new hyper-hyper parameters lol

@MarwahTAli Жыл бұрын

Thank you so much

@marbacc 3 жыл бұрын

Hello, that was great insight! Do you have code that we can look to see how to use Bayesian Search, instead of using Grid Search for Hyperparameter Tuning? Or point out good examples from someone else's code.

@geoffreyanderson4719 3 жыл бұрын

Random search iis usually a nice improvement over grid search, with minimal incremental dev complexity! And also, it just seems like a good bayes implementation would be produced by using a Thompson Sampling multi armed bandit framework! It's odd actually that we're not all doing this already by having it already provided by the popular ML framework s!

@santoshbhattarai2527 3 жыл бұрын

Waiting for a code demo!

@junkbingo4482 3 жыл бұрын

but as you pointed out, most of the people think ' i just use ml and it will do the job'...... they don't realize that in a linear model, you have variables you can select or not ( eg via tests), and no other eal parameter, but once you use non linear models ( especially if they are deep), things are different! well, as they are hired as ' datascientist', which means ' a programmer developping classes in python and/or keras and/or tf / and/or hadoop...aso', things are ok, HR are pleased sad world...