CrabNet sometimes ignores/skips certain compounds. Why? How to keep track of compound IDs?

[sgbaird]

Say there are 10,000 validation compounds and only ~9900 get pushed to the validation results. First, is this likely because of repeated compounds or certain compounds being "invalid"? Second, how to keep track of the "ID" of each compound? (in my case a Materials Project task_id).

[anthony-wang]

It's most likely due to single-element compounds which are dropped by default when loading the featurized EDM. You can find the code: https://github.com/anthony-wang/CrabNet/blob/master/utils/utils.py#L453

Also, you can check if you have duplicates in the formulae: https://github.com/anthony-wang/CrabNet/blob/master/utils/composition.py#L206

The generate_features function returns a list of skipped formulae, so you can look at that too to see what's skipped.

I'm not sure how to keep track of the MP task_id of the compound, I think that would depend on how you are storing them. Why not store and keep them separately?

[sgbaird]

Found a workaround via https://stackoverflow.com/questions/22407798/how-to-reset-a-dataframes-indexes-for-all-groups-in-one-step
related to

CrabNet/utils/utils.py

Line 455 in 9e0d79c

df = df.groupby(by='formula').mean().reset_index() # mean of duplicates

df = df.groupby(by='formula', as_index=False).mean()  # mean of duplicates

Though I'm not sure if this breaks anything

[sgbaird]

Actually, maybe something more like this is what I should be looking for:

df = (
    df.reset_index()
    .groupby(by="formula")
    .agg({"index": lambda x: tuple(x), "target": "mean"})
    .reset_index()
)

https://stackoverflow.com/questions/49216357/how-to-keep-original-index-of-a-dataframe-after-groupby-2-columns

[sgbaird]

Not sure how to factor dropping pure elements into this, other than by simply dropping them without tracking anything. Maybe by modifying groupby args.