Thoughts and to-do's on model metadata - Githubissues

role-model / roleR

R package implementing the RoLE model

https://role-model.github.io/roleR

GNU General Public License v3.0

1 stars 2 forks source link

Thoughts and to-do's on model metadata #6

Open ajrominger opened 3 years ago

ajrominger commented 3 years ago

items for metadata

includes real data or no
parameters
- parameters that are fixed versus not
- for parameters with ranges, how are they sampled across support
- model scale
data
- model/data scale
- timeseries/temporal scale
- phylogeny
- use metadata standards that exist for other data
machine learning metadata

ideas

mike has the knowledge of a consortium to standardize pop gen sim: https://elifesciences.org/articles/54967
could we draw inspiration from snakemake: https://snakemake.readthedocs.io/en/stable/
or inspiration from luigi workflow
or this: https://airflow.apache.org/

ajrominger commented 3 years ago

Metadata needs to be simple and auto-generated
ideally all we want to expect of the user is to supply a name for their analysis
house the metadata natively in R as slots in S4
make an exporter to something more language agnostic like JSON, config, YAML

ajrominger commented 3 years ago

what to do about data

easy to save the summary statistics so we should do that
question remains as to whether raw data should be stored with the analysis
- at very least, we should make a helper function that scrapes relevant metadata from raw data (e.g. spatial and temporal scope, names of files)
- give users the option to keep properly formatted raw data