pzimbrod/ml-for-phasefield

ml-for-phasefield's Issues

Training of Conv Laplacian won't converge

As discussed, the simple (3,3) Conv stencil to mimick the second-order 2D laplacian won't produce the desired stencil coefficients, regardless of optimizer, training strategy and training duration.

Normally, with an amount of parameters this small, getting the coefficients shouldn't be much of an issue. However, the optimizer doesn't seem to be able to update the stencil coefficients at all (in a meaningful manner), given the setup of the input data.

As this is the baseline for further experiments, this needs to be fixed.

missing empty foulder: data/raw

When running:
make data
within my local rep produced the stated error below. Manually creating the foulder fixed the problem.

Try 'make_dataset.py --help' for help.

Error: Invalid value for 'INPUT_FILEPATH': Path 'data/raw' does not exist.
make: *** [data] Error 2

Find a suitable baseline problem

We need a solid foundation in form of a script that covers the inference of (material) properties from an inverse problem.

For starters, looking into the DeepXDE example problems is probably a good idea