actually, he shouldn't be a professor because professors are usually horrible at "teaching". THey lecture, and profess, and do research. I'm currently taking an MIT 12-week data science course....and it's taught horribly.
At school, mathematics for me was just a set of numbers without meaning, I did not like it and did not hate it, math was just nonsense for me, but now when I was interested and began to watch videos on this topic, I realized that in fact, math is essentially a numerical description of everything, and in fact it is very interesting and not at all meaningless. I think if I had been teached that way at school, I would have chosen a technical profession
Extremely helpful videos, explained very beautifully. Would you expand this series to show math behind more complex models (decision trees, KNN, K- means) I think you would make an excellent teacher!!!
Doesn't that mean points below the fitted curve are valued differently than points above it, and affect R Squared differently? Or at least at different strenth.
Your notation is wrong. SSR is sum of squares regression and SSE is sum of square errors/residuals. Math works out, but calling shit however you like us going to confuse a lot of people.
I have two questions! 1) Why are we saying it as R^2 ( why not R)? Is it for historical reasons? 2) SST is not equal to SSR+SSE (except for special cases) because we are dealing with squares here. Then how SSR/SST represents the percentage of unexplained data? In other words (SSE/SST)+(SSR/SST) != 1 (except for special cases).
I guess that is because by definition it is equal to: sum of (yhat-ybar)^2/sum of(y-ybar)^2. As you can see we care more about variance magnitude in formula!
indeed very well explained .. just one thing (but maybe I'm wrong) ... beneath the orange data , I see n ... I think you need to divide that by n-1 ... it's a very common mistake ... there are n-1 degrees of freedom .. the summation of (xi - mean) = zero .. that means that the value (xnth - mean) 'depends' on the others to get zero in total, so it's not a degree of freedom ... this amount of degrees of freedom does not disappear even when we square the differences ...
Great video again. Just did some search why R2 is so called. Not exactly sure if my understanding is correct: R comes from Pearson's Correlation Coefficient, 2 comes from the squared from SSR/SST? Then why R for coefficient? Because it was used by greek letter rho. Then roman letter R.
It depends on how you define SSE. I've seen many books use SSE = Sum of Square Errors which is perhaps more common than the notation I use which is SSE = Sum of Square Explained. So think o my SSR as what you are probably thinking of as SSE.
You are very good at explaining, are you a professor? You should be! :D:D
He is, he just taught us something! :D
actually, he shouldn't be a professor because professors are usually horrible at "teaching". THey lecture, and profess, and do research.
I'm currently taking an MIT 12-week data science course....and it's taught horribly.
At school, mathematics for me was just a set of numbers without meaning, I did not like it and did not hate it, math was just nonsense for me, but now when I was interested and began to watch videos on this topic, I realized that in fact, math is essentially a numerical description of everything, and in fact it is very interesting and not at all meaningless. I think if I had been teached that way at school, I would have chosen a technical profession
This is the best explanation for R square I ever heard before! Thx!
Extremely helpful videos, explained very beautifully. Would you expand this series to show math behind more complex models (decision trees, KNN, K- means) I think you would make an excellent teacher!!!
Doesn't that mean points below the fitted curve are valued differently than points above it, and affect R Squared differently? Or at least at different strenth.
Crystal clear explanation! Definetly the best video on R-squared I've found on the Internet!
Excellent tutorial! Thanks for developing & sharing!
Could you please elaborate on why SSR cannot be explained by the model while SSE can be explained by the model?
Your notation is wrong. SSR is sum of squares regression and SSE is sum of square errors/residuals. Math works out, but calling shit however you like us going to confuse a lot of people.
Dude you need to open a school, you're a genius, damn bro , math has life when you teach it
I have two questions!
1) Why are we saying it as R^2 ( why not R)? Is it for historical reasons?
2) SST is not equal to SSR+SSE (except for special cases) because we are dealing with squares here. Then how SSR/SST represents the percentage of unexplained data? In other words (SSE/SST)+(SSR/SST) != 1 (except for special cases).
I guess that is because by definition it is equal to: sum of (yhat-ybar)^2/sum of(y-ybar)^2. As you can see we care more about variance magnitude in formula!
Super helpful 😃
indeed very well explained .. just one thing (but maybe I'm wrong) ... beneath the orange data , I see n ... I think you need to divide that by n-1 ... it's a very common mistake ... there are n-1 degrees of freedom .. the summation of (xi - mean) = zero .. that means that the value (xnth - mean) 'depends' on the others to get zero in total, so it's not a degree of freedom ... this amount of degrees of freedom does not disappear even when we square the differences ...
Superb explanation !
Great video again. Just did some search why R2 is so called. Not exactly sure if my understanding is correct:
R comes from Pearson's Correlation Coefficient, 2 comes from the squared from SSR/SST?
Then why R for coefficient?
Because it was used by greek letter rho. Then roman letter R.
As always Conclusions are good. thanks.
Glad you like them!
F****** amazing
Well explained. Want to know more about overfitting and it's relation with R² please. Can you provide some link to it ?
Thanks for your clear explanation. Thank you very much.
Hi, Ritvik! Your videos are being very helpful, you are very good explaining! Regards from Brazil!
Great explanation!
Glad it was helpful!
Isn't R^2 = 1 -(SSE/SST)?
It depends on how you define SSE. I've seen many books use SSE = Sum of Square Errors which is perhaps more common than the notation I use which is SSE = Sum of Square Explained. So think o my SSR as what you are probably thinking of as SSE.
Thanks for quick reply :)
Superb. Very nicely explained 🙏
you should get a bigger paper
Hi, excellent explanations. I have a question, where do you explain overfitting?
Thank you for your kind words. You can find my overfitting video here: ruclips.net/video/-JopeGg60QY/видео.html
Dude, you don't know how good you are...
Great explanation. Thank you very much!
Too fast, i must reduce the speed to understand !!!
thats why u hav that option...
you save me
Very clear explanation. Well done!
awesome! thanks :)
Great! thanks!
OK
Excellent. thanks!
Clear.
Incorrect and misleading albeit "good" explanation.
+Sarah Chen Care to elaborate?