Question: the bi-variate cost function plot with regard to w and b is a parabolic surface here.

Is this only the case when the target function is linear? What happens if the target function is non-linear and we use linear regression?

Question: the bi-variate cost function plot with regard to w and b is a parabolic surface here.

Is this only the case when the target function is linear? What happens if the target function is non-linear and we use linear regression?

The cost function for linear regression is always convex, because it is based on the sum of the squares of the errors. This curve is always parabolic.

The shape of the hypothesis does not matter in this regard.

Thatâ€™s right. Thanks! The shape of the target function when compared to the hypothesis function will only influence the curvature of the cost function, right?

It depends on what you mean by â€ścurvature of the cost functionâ€ť.

Gradient is more appropriate here, I reckon.

The gradients are the partial derivatives of the cost equation.

1 Like