Understanding Mean Absolute Error and Mean Squared Error as ML metrics and loss functions

Tips Tricks 37 - MAE vs MSE vs Huber
Understanding Mean Absolute Error and Mean Squared Error as ML metrics and loss functions

Use MSE if outliers are important.​

USE MAE if outliers are not important (most cases).​

Use Huber to get a balance between giving outliers some weight but not a lot (like in MSE). ​

[reganmian]

I'm in a graduate level statistical inference course. I only had probability in undergrad with no statistics. This was such as nice and fast explanation to give motivation to why I'm learning this. Rigor has it's place, but I needed this

[ssyedyaseens]

I am very new here.
Why is mae an array during code demo?
Isnt it supposed to be just one value?

If x is data points, and y is some predicted value, y_delta could be an array, how can y_mae be an array?
Appreciate any help. Thanks

[olaoyedamilola]

At what level is MAE and MSE is accepted to be good or bad

[brunospfc]

Thank you professor, i have a question, given a confusion matrix result of my prediction as such : 1st line [3 0 0] 2nd [0 3 0] 3rd [1 0 2], how to trace my way back to the image that generates de "1" in the third line ? So i can see if there's something particular with this image ? have a good day

[pankajgoikar]

Excellent, Everything is cleared. Thank you so much.

[Mojtaba-Sirati-Amsheh]

please, make a video about GRAD_CAM for image regression tasks

[teklehaimanotaman]

Thank you very much for your outstanding tutorials. My question is why does an outlier remain important? before we train our mdoel, we have to clean outliers and interpolate missing values. Could you please explain this please? Thank you.

[elzeardbouffier]

Leaving a comment to say that you have helped me greatly improve my research.

[aryansakhala]

Professor, please consider my resume for research with you...

[minhajulhoque]

Really enjoy your videos. Very clear and concise. Thank you!

[divyaanshsingh]

Quick question: COntradiction to your summary - If outliers are important shouldnt we use MAE instead of MSE since MSE is highly influenced by outliers?

[KN-txsd]

Sir, we now often see in some analysis people showing MAE as 0.20 ± 0.011 as well as R2 as 0.45 ± 0.028, can you kindly explain how these ± values are obtained for MAE and R2 for any analysis can any of the python packages can derive these values. Thank you