Robust Regression with the L1 Norm [Matlab]

[@LucasL.Mbembela]

Along with visual aids you have explained the concept in a very understandable manner. Thanks for the video.

[@MaksymCzech]

This is so amazing. Can you please hint at how you visualize those formulas. Do you have a glass in front of you or is it all in computer?

[@RJ-ie8cc]

Just found your channel, great videos! Thank you.

Hi, great channel! How do you display the image on your lightboard? Do you use any projector?

Or maybe you have a TV on the other side and there you can see what you are currently showing and the graphics are added live to the camera image?

[@matthewjames7513]

wow never heard of CVX before. Is that generally better than fmincon?

[@looper6394]

You meant lasso

[@Eigensteve]

fmincon is good for quick and dirty optimization; it is for constrained optimizations and I don't know if it needs to be convex or not. My understanding is that CVX is pretty highly optimized for convex optimization, and has a more formal language for how to specify the problems.

[@pipertripp]

I'm really new to this stuff so I hope that this isn't an asinine question. Can you just throw out outliers if you can justify that they're rubbish data points? Is the advantage of this approach that you aren't required to justify the removal of outliers (and all the work that would go into that)?

[@Eigensteve]

Great question -- that is exactly right. This saves a ton of time potentially, and can be implemented in real time without human intervention.

[@pipertripp]

@@Eigensteve I found the python flavour of this vid and you spoke to it a bit there. I'm teaching myself so it's nice to get some input from somebody with loads of experience.

[@abderrahmanechamsi3060]

thanks for the video one question please how can i switch the layout of Matlab to make it look like the one in the video

[@wazirakbar1218]

MATLAB--HOME-PREFERENCES-COLORS and you can change the background, comment, sections etc colours there.

[@haschka83]

Yes, I guess that L1 also becomes more popular due to the increase of available computing power. L2 still provides a fast analytical solution. L1 has to be iteratively minimized which can be way more inefficient than just analytically solving least squares (lots of data points and high dimensional fit). I found that out the hard way when I bought compute time to fit atomic partial charges to electrostatic potentials using L1.

[@Alwadah]

Could you collaborate with Robert van De Geijin and Maggie Myers at UT Austin? They have beautifully organized set of courses on Linear Algebra (Numerical Analysis with Computation). Your videos with their organization would be something else.

[@anatolystrashkevich7621]

thank you!

[@petersagitarius4356]

It looks very good, but it is not the native Matlab. Operations like cvx_begin / cvx_end is not part of original matlab language.. For my purpose it is useless :( , because, I cant instal new libraries into system... Could you suggest some alternative solution for pure matlab without installing extra libraries please ? Thnx

[@evanparshall1323]

I'd just like to comment that least squares and pseudo inverse are not the same thing