Error cost function shape

[KiraDiShira]

In the following slide we learn that squared error cost is not always convex:

I don’t understand the answer in this thread:

Squared error Cost function shape

I just wanted to know is the of squared error cost function always convex (shape of a bowl) Or that applies only for linear regression?

So I have a question. For linear regression squared error cost function is always convex. Can we say the same for the polynomial regression we saw in week 1?

And we can say that the cost function that we learn in week 2:

is convex also when we choose non-linear decision boundaries?

Thank you

[TMosh]

My reply applied to linear regression - not logistic.

The logistic cost function is also known to be convex, because it doesn’t use the squared error.

[TMosh]

@KiraDiShira

Can we say the same for the polynomial regression we saw in week 1?

Yes, we can say that. The cost is the sum of the squared errors - it does not matter how the features are created.