Well, although you made a really good and concise video, I think when we say RMSE penalizes outliers, it means opposite - it is less robust to outliers than MAPE. If you look at the formula, it is indeed squared of errors and hence, when it comes to large error, it will be squared. Penalizing outliers simple means " RMSE tends to give more weight to outliers" so that large deviations tend to be more visibie. I generally agree that RMSE would be better choice for hyperparameter tuning when your focus is minimizing absolute error and also to prioritize reducing the impact of larger errors (potentially at the expense of smaller errors), then RMSE would be your choice and it would be most of scenarios. If your primary goal is to minimize the relative percentage error in predictions and you want to prioritize small errors across the entire range of actual values, MAPE might be a better choice. And well, you would go with RMSE first and then if you can't really find good hyper-parameter to minimize it, then go with MAPE, for instance. Or, depending on the problem, you may go with MAPE first.
Nice video, but I think you make an error with the outliers: RMSE is the std.dev. of the error and highly affected by outliers (as std.dev. always is). MAPE is more related to the median and therefore more robust to outliers. So exactly the other way round.
Please help me with this question's answer... How to convincingly answer in interview for having a long career gap for 4 years and now joining into data science? How to answer?
Well, although you made a really good and concise video, I think when we say RMSE penalizes outliers, it means opposite - it is less robust to outliers than MAPE. If you look at the formula, it is indeed squared of errors and hence, when it comes to large error, it will be squared. Penalizing outliers simple means " RMSE tends to give more weight to outliers" so that large deviations tend to be more visibie. I generally agree that RMSE would be better choice for hyperparameter tuning when your focus is minimizing absolute error and also to prioritize reducing the impact of larger errors (potentially at the expense of smaller errors), then RMSE would be your choice and it would be most of scenarios. If your primary goal is to minimize the relative percentage error in predictions and you want to prioritize small errors across the entire range of actual values, MAPE might be a better choice. And well, you would go with RMSE first and then if you can't really find good hyper-parameter to minimize it, then go with MAPE, for instance. Or, depending on the problem, you may go with MAPE first.
Nice video, but I think you make an error with the outliers: RMSE is the std.dev. of the error and highly affected by outliers (as std.dev. always is). MAPE is more related to the median and therefore more robust to outliers. So exactly the other way round.
thank you for the video. Please keep posting your case and easy explaining on your website. I am on it.
Why did you stop videos ???...You need to continue man !!!!
Hi Pavan!
Thank you for the encouragement!
Yes, I will be posting more videos soon.
@@thinking_neuron waiting
your videos are nice and practical,please continue
Hi Rafi
Thank you for your kind words!
I will post more videos soon!
@@thinking_neuron ,Hi bro please start doing videos ,if possible please share your email id so I can contact you
Please help me with this question's answer...
How to convincingly answer in interview for having a long career gap for 4 years and now joining into data science?
How to answer?
Hi Aashqeen!
Please go thru the below list of videos.
thinkingneuron.com/data-science-interview-final-checklist/
Got it the difference