Suggestion: `MappingSetDataFrame`: `.to_csv()` and `read_csv()`

[joeflack4]

Overview

I was initially suggesting that we could have an alternative constructor for MappingSetDataFrame that could accept paths rather than objects, however, based on some more recent discussions (1, 2), it might be better to do design very similar to pandas:

Reading

msdf: MappingSetDataFrame = read_csv()

Or read_tsv(), or just read().

It could accept various different combinations of params:
a. Path to a SSSOM TSV which includes a metadata comment.
b. Path to a SSSOM TSV with no metadata comment, and path to a metadata yaml.
c. Path to a SSSOM TSV with no metadata comment, and path to a metadata object.
d. a dataframe, and path to a metadata yaml.
e. a dataframe, and path to a metadata object.

Writing

msdf.to_csv()

Or to_tsv().

This would be an alternative or replacement to parse_sssom_table().

[matentzn]

I see the point, but its a bit of a lower priority. I would probably recommend a utility method

def load_mapping_set_from_df(str: mapping_set_path, str: metadata_path):
        ...

Rather than adding further IO logic into the MappingSetDataFrame class.

[joeflack4]

That would be fine, but I'm not sure if I like it either. Another thing that a user needs to know to import.

The logic for MappingSetDataFrame if it were a normal class would be a few lines, but I just realized that I don't know how to handle this for data class setup(s). And I guess with LinkML, it has to be a data class.

[matentzn]

I would say we go with the pandas design, as it is familiar to other devs, i.e. pd.load_csv(..)

[joeflack4]

Yes, I like that solution!

[joeflack4]

I just changed the title and OP of this issue from being about allowing MappingSetDataFrame to accept paths, to instead being about pandas-like functions/methods.

[cthoyt]

#589 adds an alias sssom.parse_tsv, which partially addresses this. It also adds an alias sssom.write_tsv(), so you can do:

import sssom

# other SSSOM files can be found on https://mapping-commons.github.io
url = "https://raw.githubusercontent.com/mapping-commons/mh_mapping_initiative/master/mappings/mp_hp_eye_impc.sssom.tsv"

# TSV can be parsed into a mapping set dataframe object,
# which includes a pandas DataFrame, a curies.Converter,
# and metadata
msdf = sssom.parse_tsv(url)

# SSSOM comes with several "write" functions
sssom.write_tsv(msdf, "test.tsv")
sssom.write_json(msdf, "test.json")
sssom.write_owl(msdf, "test.owl")
sssom.write_rdf(msdf, "test.ttl")