battler. wrote:
For some reason providing the partial derivatives produces worse (this is based on the results I have from the gradient descent method) answers.
I'm sure the partial derivatives have been derived correctly.
If you are certain the partial derivatives have been derived correctly I would double check that they have been implemented correctly. Each one should have its own while loop for example.
I was dubious so I added the partials to my fit and got no difference from the results using finite difference approximations.
