Unique Atom Types and Charges

[shenoynikhil]

For each dataset, it would be nice to have a property that returns,

• Unique Atom Numbers
• Unique Atom Number - Charge Pairings
• Mean and Std of energies

[prtos]

@FNTwin add meta statistics about each dataset and the full collection. statistics must include

• number of unique molecules
• number of unique molecules and charge
• number of conformers per molecules distributions
• heatmap of number of share molecules between datasets
• energy distributions per conformers, energy range distributions per molecule
• input distributions: chemical space, conformational space
• heatmap of similarity of input spaces of between dataset.
• heatmap of similarity of target (energy and force) spaces of between dataset.

[shenoynikhil]

@prtos Just as a heads up because this might be an issue during this task, some datasets like ANI and Nabla have the len(data['name']['uniques']) == number of conformers and not number of molecules. I checked them and it seems like it is a unique identifier for a conformer and not a molecule.

[prtos]

ok I see. I'll look into it.