Statistical Learning: 2.2 Dimensionality and Structured Models

[@MrWater2]

Please improve the quality of image because the material is extremely good

[@ashleyforte2885]

Love the topological breakdown.

[@Tsicloh]

I feel like the approach in the book and that of these videos is totally different.😅

[@ashfaqmahmudshovon9097]

It seems so...

[@Sama-mn3hh]

For the curse of dimensionality, can't we also use dimensionality reduction?

[@brucepackard2729]

Income follows a power law, pareto distribution though. What happens when you include a billionaire who didn't finish university education as a datapoint? You could discard the billionaire as an outlier, however you lose a lot of representativeness of real life if you do that.

[@Sama-mn3hh]

In practice, you use the z-score remove outliers from the dataset as a part of data prep before fitting a model. If the model is focused on predicting the income of the working class, then that would be a solution. Depends on the scope of the project I suppose.

[@PaulMugo-r9c]

Why do you use a circle when getting the nearest neighbor in 2 dimensions

[@mattetor6726]

The first neighborhood is just using X1, ignoring X2. When you also take X2 into account, you start at the point (x1, x2)( =(0,0) in this case) and spread out with a circle until 10% of the data points are covered.

[@billsurrette6092]

You don’t have to use a circle. What you must so is define how you’re going to measure “distance” when determining how far away the points are. When you use Euclidean distance you get a circle. But there are other distance measures: Minkowski, Manhattan, etc. These will create other shapes, not circles. So the circle is the byproduct of the important thing: your chosen distance metric.

[@PaulMugo-r9c]

Thank you all for replying! I understand

[@yulu-520c]

do you guys feel confused about everything

[@yulu-520c]

what is p

[@yulu-520c]

oh it is the number of variables, no problems now

[@anthronox4992]

You should read the book, too. It's explained in more detail there (and more easily, imo).

[@yulu-520c]

@@anthronox4992 get it, thx