Create LinkML files for all Toy data

[twhetzel]

Given the toy data in toy_data/initial of:

• demographics.tsv
• lab_results.tsv
• sample.tsv
• study.tsv
• subject.tsv

Create LinkML using SchemaAutomator for each file. It is ok to have each of these as separate model files.
The documentation for SchemaAutomator should be updated with any new information needed.

[twhetzel]

Based on the documentation, there are two possible commands:

• Run on each data file individually using generalize-tsv as:

schemauto generalize-tsv --schema-name Demographics ../initial/demographics.tsv -o Demographics.yml

and

• Run on all data files using generalize-tsvs as:

schemauto generalize-tsvs --schema-name Demographics ../initial/demographics.tsv --schema-name LabResults ../initial/lab_results.tsv --schema-name Sample ../initial/sample.tsv --schema-name Study ../initial/study.tsv --schema-name Subject ../initial/subject.tsv -o toy_data-all.yml

[amc-corey-cox]

Trish, I haven't looked deeply into this so you likely know more than I do but I was able to build it all into one schema using this:

schemauto generalize-tsvs ../dm-bip/toy_data/initial/*

Now, I don't know if that is a good schema or if there are reasons we would want this to be a separate schema for each file but my sense is that having it all as one single schema is probably more flexible for different data sets so we don't have to rely on manually specifying schema names.

[amc-corey-cox]

We do probably want to name the schema when we make it... like this.

schemauto generalize-tsvs --schema-name Toy_Schema ../dm-bip/toy_data/initial/*

[twhetzel]

Yes, I imagine that schemauto generalize-tsvs ../dm-bip/toy_data/initial/* also works, but gather the schema-name arg might be useful. There are some discrepancies between the web docs, cli docs, and what commands actually work so I've been trying out different things to try to understand what works and how commands are intended to be used.

Having one model file is fine with me and a question I wanted to run by you.

[amc-corey-cox]

What I try to keep in mind is that ideally we'll have essentially no human interaction in this. So we want to be able to put the data somewhere, target that location and have it do everything, perhaps with a variable to say what dataset we're working on.

[twhetzel]

for the conversion, yes, no human interaction, but then a human will need to review the file(s) that are generated