Add tests to ensure quality #27
Description
We should do two kinds of tests:
Sanity checks
- Basic checks to avoid a faulty system to go online. This should include:
- Unit tests for common inputs / functionality (hit predict with empty input, invalid SMILES, ..., upload, download, insights)
- Check that models are producing realistic results - for some cases, check that some obvious classes are predicted / not predicted (this should uncover errors like swapping the classes.txt file)
Quality checks ("Competency questions")
- A more extensive suite of molecules and correct classifications for them (even ones we cant answer at the moment)
- This should be used to track progress in subsequent iterations