Add neural-net parameter tests

[sebapersson]

Relatively big update adding more tests for neural network parameters, addressing the parameter points in #11. Specifically, new tests include:

• Having neural network parameters which are not estimated (Hybrid tests 18–19).
• Specification of nominal values for layers and/or parameter arrays (Initialization tests 1–3). These tests verify that the parameter table can specify parameters for a specific layer.
• initializationPriorType sampling for neural networks (Initialization tests 4–16). These tests confirm that initial value sampling works for glorotUniform, glorotNormal, kaimingUniform, kaimingNormal, normal, and uniform. They also verify that different layers or parameter arrays can have different initializationPriorType settings (e.g., needed because the gain for kaimingNormal depends on the activation function). For more details, see the specification.

Additionally, for all hybrid tests, the parameter file with nominal values is now specified in the parameters table. Lastly, I’ve reorganized the test_cases directory by splitting the three test types into separate directories. Everything is tested against PEtab.jl as usual.

If all this sounds good @FFroehlich and @dilpath I will merge this.

[FFroehlich]

Sounds good to me

[dilpath]

Sounds good to me too!

Is hf5 a common extension for HDF5 files? I am used to h5 or hdf5 (e.g. see https://en.wikipedia.org/wiki/Hierarchical_Data_Format )

splitting the three test types into separate directories

Looks much nicer. Should these folders be deleted?

• 16 folders matching test_cases/{001..016}
• 53 foldesr matching test_cases/net_{001..053}

[sebapersson]

Thanks for the feedback!

Is hf5 a common extension for HDF5 files? I am used to h5 or hdf5

This is a mistake on my part, I will change to .hdf5 ending as I think it is more clear.

Looks much nicer. Should these folders be deleted?

Seems I forgot to use git rm -r :) I will fix.