add an `ExcludeFilter`

[mathause]

In my IPCC analyses I had to remove many simulations (model-variable-, model-scenario-ensemble-, or other combinations). Even for the cleaned 'new-generation' repos, mesmer has to remove some of the simulations. For IPCC I did that in the data processing loop but I think it would be better done in the 'find the simulations' part (i.e. in filefinder). So we should add an ExcludeFilter (better names always welcome). We'd need to think about the way metadata for the excluded simulations is passed.

For IPCC I have a function which identifies matching metadata:

https://github.com/IPCC-WG1/Chapter-11/blob/d1a3a99f242a568fb4cefc36a038c888a90b9d37/code/fixes/_fixes_common.py#L47

But maybe could also use pandas machinery, e.g. isin():

# NOTE: untested

conditions = [
    # remove AWI ocean data: has an unstructured grid
    {
        "table": ["Oday", "Ofx", "Omon", "SIday", "SImon"],
        "model": ["AWI-CM-1-1-MR", "AWI-ESM-1-1-LR"],
    },
    # tasmax and tasmin are wrong for CESM
    {
        "table": "day",
        "varn": ["tasmax", "tasmin"],
        "model": ["CESM2", "CESM2-WACCM"],
    },
    ...
]


to_keep = True

for condition in conditions:
    to_keep |= ~ all(df[key].isin(cond) for key, cond in condition.items()))


df = df.iloc[to_keep]

FYI @veni-vidi-vici-dormivi

[veni-vidi-vici-dormivi]

That is a useful idea! As far as I can see, isin can take dictionaries, so even

for condition in conditions:
   to_keep |= ~all(df.isin(cond))

could work.

I like ExcludeFilter. Also find_files could have an exclude= parameter but it might become too bulky...