Week 3 linear model

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

Why in the video, it says " The notion of high bias and high variance, it doesn’t really happen for linear models applied to 1D"?

image
image1920×1009 112 KB

2

Personally, I have not seen a model that has both high variance and high bias.
It is difficult to invent an example of when that would occur.

3

If the truth is a curve and we use a straight line to predict it, then the result will have high bias and high variance, because the training set error and cv set error will both be high. No?

4

That’s an example of high bias, but not high variance.

High variance is characterized by very low training cost. A straight line is not going to give low cost if the data is a curve.

5

Example:
Take three points that are derived from the equation for a parabola. That’s the training set.

Now make a validation set using one additional point, that does not lie on the parabola.

  • Since a parabola can always be fit to three points, the training cost will be zero.

  • Since the validation set is not on the parabola, the validation cost will be high.