Machine Learning 1 - Linear Classifiers, SGD | Stanford CS221: AI (Autumn 2019)

[@pavittarx]

I am fiinding this course to be very interesting and answering a very fundamental questions I have had for a long time. I'm loving this approach as compared to everyone running into types of models etc, whenever they discuss AL/ML.

[@swavekbu4959]

Excellent video and teacher. The "weight vector" is nothing more than a linear combination, where each feature is weighted by a scalar that is estimated. The goal is to estimate an optimal linear combination of features that will maximally separate groups if it's a linear classifier. E.g., suppose you want to classify who will make their monthly mortgage payment and who will not. What are some good features (predictors) to use to classify? You will choose these based on theory or what you think will work. Suppose yearly earnings is one such feature, and past payment behavior is another. We seek a linear combination of yearly earnings and past payment behavior that optimally separates the binary response variable. This is what a classical linear classifier attempts to do. The key point is that a "weight vector" in this case is just a linear combination of variables, weighted in such a way as to optimize classification on the response variable. As a comparison, ordinary least squares regression seeks a weighted linear combination such that error is minimized in prediction. Both concepts are very similar, they just apply to different type of responses (binary vs. continuous).

[@leventaksakal5]

No laughs at 33:33 come on guys

[@WhiskeyTangoFoxtraught]

Thank you for posting. No Lecture 1, did I miss anything important?

[@AnmolKumar-dh3lh]

interesting thank you so much!!

[@urduadab90DrEmraanAzfar]

I register myself for the learning of Artificial intelligence but there is Math & Math all around. what i can do?

[@qasimmalik6560]

unregister? :)

[@YeekyYeeky]

really great class

[@Jonathan-ru9zl]

Where can I get the sources?

[@annorprince6321]

Please, I really love soccer because I used to play soccer. So I'm wondering if I can become Data Scientist in soccer by learning Data Science?

[@Arne_Boeses]

Yes of course!

[@kiwi-mf2do]

Soccer definitely handles a lot of data, the stats of the players teams etc. So yes.

[@Hiphiphooray490]

Sure! But please note that learning the rules of soccer is a bit easier than learning data science.

[@alexplaytop]

All about math, but if I want to train chatbot like app? Where only words or data of strings?

[@oscarcoca5385]

Bro this is class 1, chill. Strings can be treated as numerical vectors

[@gertrudevine9678]

Which software language is the tutor using for his coding?

[@ousmanedu3694]

python

[@SimsonPoon]

Why phi(x) is an array in each point value?😮‍💨

[@yagmurgulec1754]

How come the vector [2 -1] can be perpendicular to the vector [2 4]? Instead, maybe you could've meant [1 2].

[@notintheobservableuniverse2594]

Their dot product is apparently 0, they should be perpendicular (you're probably confusing "perpendicular" with "parallel").

[@yagmurgulec1754]

When they are perpendicular to each other, their dot product should be zero.

[@sourabhverma9034]

Slope = y/x. So vector [2, -1] slope = -.5 vector [2, 4] = 2 2 lines are perpendicular if dot product of their slope is -1 See 2 * -0.5 = -1

[@fishfrompluto]

They are perpendicular and dot product is 0. In algebra the dot product of points (a1, b1) and (a2, b2) is a1*a2+b1*b2 For points (2,-1) and (2,4) , the dot product is 0.

[@monkmode9138]

It's too theoretical lecture to me, i am suffering.

[@Eizengoldt]

Rip