materials-science

As it says, move away from keeping n_bands fixed during the SCF.

Per SVK's suggestion, it is worth exploring if one can use an autoencoder (regular or variational) instead of PCA for the sliding-window-based image denoising originally written by @ramav87 .

@saimani5 @markpoxley will you be interested in exploring it?

It would be a bit of fun to see if stk can handle building things like this: https://chemrxiv.org/articles/preprint/Isogonal_Weavings_on_the_Sphere_Knots_Links_Polycatenanes/12616268

Summary of discussion with @utf and @scanlond based on user feedback.

• Users can get confusing/incorrect results when using kgen with a non-primitive POSCAR
• kgen warns that results may be incorrect and writes a POSCAR_prim file along with a KPOINTS which corresponds to that primitive/standard cell setting.
• If the user ignores this warning and uses their current POSCAR with this

add a section on hyperparameter tuning since classical models were used with default hyperparameters

The semantic segmentation models need to be extended to work with 3D data. This should be very straightforward - just introduce an option to select between 1D, 2D, and 3D cases to ConvBlock, [UpsampleBlock](https://g