8

Question 1

Why reduce dimensionality and what are the drawbacks

Answer 1

To speed up training
To visualize data
Compression

Information is lost
It can be intensive
It adds complexity
Transformed features are hard to interpret

Question 2

Whats the curse of dimensionality

Answer 2

The more dimensions the more likely over fitting is
The harder it is to identify patterns
More training data is required

Question 3

Can dimensionality reduction be reversed

Answer 3

Its impossible for full reconstruction - however sometimes an attempt can be made

Question 4

Can PCA reduce dimensionality of a non linear data set

Answer 4

Yes as it can get rid of useless dimensions however if there are no useless dimensions then it can’t be done

Question 5

How can you evaluate performance of dimensionality reduction

Answer 5

Apply the reverse then measure reconstruction error

Question 6

Should you chain dimensionality reduction algorithms

Answer 6

Yes this is often useful as it may save time