Hi, First of all, thanks for the feed-back! You are totally right that in some slides there is too much text, and sometimes it is difficult to read. Think that these slides were prepared for teaching at class. We are working in versions with more dynamic plots and more understandable. Cheers Q&T
Just to clarify, the "loadings" you used, such as BNP and infant death rates, are arbitrary scores (I would use the basis vector or a value of 1) assigned to a specific attribute and 0 for the other attributes [i.e. BNP would be (1,0,0,0,0,0,0) and infant death rates can be (0,1,0,0,0,0,0), and so on, in the original 7-dimensional plot]. When you make the bi-plot or the 2-dimensional covariance plot, you are essentially projecting the coordinates in the 7th dimension to the plane of most variation (1st and 2nd principle components). Now, your "loadings" in the 2-dimension then reflect the projections from axes in the 7th dimension, so you can roughly see how your samples relate to each axis in the 7th dimensional space. If none of above makes sense or is incorrectly, my question is: where are you getting the scores to plot the "loadings" onto your bi-plot
The loadings are not the original basis vectors. We are seeking a new set of basis vectors (loadings) that can describe as much as possible of the variation in as few dimensions as possible. If you check the following videos, this will be explained in more detail. You can get more formal explanations at en.wikipedia.org/wiki/Principal_component_analysis.
Thank you so much for these videos. They are clearly explained and pace of learning is composed. Great examples that learners can relate to. Thank You!
Great videos! Thanks for putting these up, they helped me a lot.
11 лет назад
I actually don't mind the slides at all. I prefer watching the videos at 1.5 speed and pausing when I need to. That's the great part about RUclips, that you can control your own speed.
Nice projection of the unnecessary details and sorry for being a bit hard on you but the approach is a bit of everything and not essential explanation - introduction about PCA
Thanks for all the positive comments. Those are really appreciated.
Hi,
First of all, thanks for the feed-back! You are totally right that in some slides there is too much text, and sometimes it is difficult to read. Think that these slides were prepared for teaching at class. We are working in versions with more dynamic plots and more understandable.
Cheers
Q&T
i wish my prof was as clear as you - he just throws formulas at us with no visual explanations thank you !
Thanks so much Evangelina. I am truly glad the videos are helpful.
Prof Bro, your explanations make a concept such as PCA appear so simple. Thank you.
Just to clarify, the "loadings" you used, such as BNP and infant death rates, are arbitrary scores (I would use the basis vector or a value of 1) assigned to a specific attribute and 0 for the other attributes [i.e. BNP would be (1,0,0,0,0,0,0) and infant death rates can be (0,1,0,0,0,0,0), and so on, in the original 7-dimensional plot]. When you make the bi-plot or the 2-dimensional covariance plot, you are essentially projecting the coordinates in the 7th dimension to the plane of most variation (1st and 2nd principle components). Now, your "loadings" in the 2-dimension then reflect the projections from axes in the 7th dimension, so you can roughly see how your samples relate to each axis in the 7th dimensional space.
If none of above makes sense or is incorrectly, my question is: where are you getting the scores to plot the "loadings" onto your bi-plot
The loadings are not the original basis vectors. We are seeking a new set of basis vectors (loadings) that can describe as much as possible of the variation in as few dimensions as possible. If you check the following videos, this will be explained in more detail. You can get more formal explanations at en.wikipedia.org/wiki/Principal_component_analysis.
Rasmus Bro Thanks for the reply! Great video, by the way
Thank you so much for these videos. They are clearly explained and pace of learning is composed. Great examples that learners can relate to. Thank You!
Thanks, this has been one of the best explanations Ive received. This topic might be rather complex but you made it quite clear and nicely explained.
Thanks so much for your comment. We are really happy that you find it useful
A really informative and well explained pair of videos! Just what I needed! Cheers!
Thanks putting this up. It helped me a lot in finally understanding PCA and the potentials!
Great videos! Thanks for putting these up, they helped me a lot.
I actually don't mind the slides at all. I prefer watching the videos at 1.5 speed and pausing when I need to. That's the great part about RUclips, that you can control your own speed.
You are a very good lecturer.
This is wonderfully helpful, thank you
This video helped me a lot. Thank you.
thank you very much I actually understood PCA....
Excellent explanation.
Thank you. I found it very useful.
excellent video ! Thanks so much, really helped me :)
thank you very much. I find it extremely helpful :)
What software did you use to generate these plots?
Nice, really nice!
Thank you very much
Great Great Great. thank you
What are these plots? What's on each axis?
These are score, loading and bi-plots.
Most of the plots are made in MATLAB and are then modified in Powerpoint.
thank u
Nice projection of the unnecessary details and sorry for being a bit hard on you but the approach is a bit of everything and not essential explanation - introduction about PCA
No problem. A lot of people tend to disagree, but you can never make everyone happy. So no worries at all :-)