feat(nmrcraft and scripts): baseline experiments

[mlederbauer]

always min. 6 tensors as input 1st: 6 tensors -> metal category 2nd: 6 tensors -> E-ligand 3rd: 6 tensots -> metal, E-ligand category 4th: 6 tensors, metal, E-ligand, 5 ligands -> 6th ligand

generate for each of those tasks:

• plots and scores accuracy, ROC, ...
• different dataset sizes: 1%, 10%, 100%
• anything else?

[mlederbauer]

It would be fantastic to prioritize this to be able to get results until Friday @wikicraft @mlederbauer @tiaguinho-code @strsamue @kbiniek

[mlederbauer]

We should be able to adapt the DataLoader class “dynamically” and then select the target and feature columns depending on which experiment we want to run see PR #31

[tiaguinho-code]

single target experiments should be doable now with the DataLoader. For multitarget experiments we still need to add onehot encoding, as the current dataset we get is categorical.

[mlederbauer]

idea: extrapolating by training on 6 coordinate, testing on 5-coordinate or vice versa

[mlederbauer]

1.) input: 6 NMR tensors, output: metal 2.) input: 6 NMR tensors, output: E-ligand 3.) input: 6 NMR tensors, output: X3-ligand 4.) input: 6 NMR tensors, output: metal & E-ligand 5.) input: 6 NMR tensors, output: metal & X3-ligand 6.) input: 6 NMR tensors, output: E-ligand & X3-ligand 7.) input: 6 NMR tensors, output: metal & E-ligand & X3-ligand

8.) input: 6 NMR tensors+all other ligands, output: metal 9.) input: 6 NMR tensors+all other ligands+metal, output: E-ligand 10.) input: 6 NMR tensors+all other ligands+metal, output: X3-ligand 11.) input: 6 NMR tensors+all other ligands, output: metal & E-ligand 12.) input: 6 NMR tensors+all other ligands, output: metal & X3-ligand 13.) input: 6 NMR tensors+all other ligands, output: E-ligand & X3-ligand 14.) input: 6 NMR tensors+all other ligands, output: metal & E-ligand & X3-ligand

those are quite a lot, I suggest to keep the following, what do you think?

Research Questons:

• Does more data = better results?