`write_table()` can ignore `curie_map` in `msdf.metadata`

[joeflack4]

Overview

I'm creating a MappingSetDataFrame (msdf) and then saving it:

    msdf = MappingSetDataFrame(matches_df, metadata={
        'curie_map': {
            'rdfs': 'http://www.w3.org/2000/01/rdf-schema#'
        },
    })
    with open(outpath, 'w') as f:
        write_table(msdf, f)

When I inspect msdf, I see that it does have my correct curie_map set inside the metadata property.

However, when I write_table() and inspect the resulting TSV, the curie_map is empty within the metadata comments at the top of the file:

# curie_map: {}

Update: This problem does not happen when instantiating an msdf using from_sssom_dataframe().

[matentzn]

Can you try a non-standard prefix? like LOINC or some such?

[joeflack4]

Same problem. Both of these attempts result in the same empty curie_map:

    msdf = MappingSetDataFrame(matches_df, metadata={
        'curie_map': {
            'abc': 'http://example.org/abc#',
        },
    })

    msdf = MappingSetDataFrame(matches_df, metadata={
        'curie_map': {
            'abc': 'http://example.org/abc#',
            'rdfs': 'http://www.w3.org/2000/01/rdf-schema#'
        },
    })

[matentzn]

I think the problem is that you have to use the parse_* methods.. the mappingsetdataframe constructor should really be private, or the whole MSDF class refactored to act more like a pandas dataframe..

sssom-py/src/sssom/parsers.py

Line 281 in e2f67de

def parse_sssom_table(

[joeflack4]

Mm, well firstly, I don't know if there is a good reason / convention for keeping the constructor private. I think it's a good thing to have object constructors, and it makes sense in this case. Straightforward and easy to use.

Unfortunately the parse_*() methods do not support instantiation from an existing pandas.DataFrame, which is my use case (and I have had several such use cases in the past). They are for reading from disk.

Also same problem with these parse_*() methods; only 1 of them shows up in the Python API documentation.

[matentzn]

Unfortunately the parse_*() methods do not support instantiation from an existing pandas.DataFrame, which is my use case (and I have had several such use cases in the past). They are for reading from disk.

from_sssom_dataframe(df...)

Is what you need I think, like:

msdf = from_sssom_dataframe(df, prefix_map=converter, meta=combine_meta)

[joeflack4]

Ah OK. This probably escaped my eye because I read "from SSSOM dataframe" as meaning that I will be passing it something like an msdf rather than a normal df. Not sure how else I'd name it though.

The Python API docs are in need of a big update, as these from_*() methods are also missing.

This is indeed what I need. It does not have the same curie_map issue when write_table() is used. It also adds validation.

There is a problem where it ignores prior sorting of the df that is passed in; Opened a new issue about that:

• from_sssom_dataframe() ignores prior sorting #574