Model type inference

[brandomr]

There are many types of models: physics based models, ODEs, machine learning models, deep learning, etc.

Can we analyze the model imports and how they are used to determine the type of model?

For example, a model that imports pytorch is likely doing some deep learning.

A model using odeint is probably doing something with ODEs.

[GoogleSheets]

Do you see this inferring a single identified model type, or would it be more useful to categorize imports into several bins e.g. ODE, Fourier Transforms, Linear Algebra, to summarize more specifically what the model does?

Also for example, instead of just deep learning for the pytorch libraries, specify audio, text, image/video libraries.

single classfication vs. multiple classifiers

[brandomr]

I think a model could be assigned multiple categories. Subcategories like "multi-class classifier" would make sense as well, but in the World Modelers context I think we are more likely to see physics/theory based models than ML ones. I'm less sure how to detect and categorize those.

Here's an example of a hydrology model for context: https://github.com/peckhams/topoflow36

Here are the largely ML based kimetrica models: https://gitlab.com/kimetrica/darpa/darpa/-/tree/master/models

Here is another hydrology model: https://github.com/PSUmodeling/MM-PIHM

Would be interesting to see what Flee is...

[GoogleSheets]

Just pushed update using data_files/ model-type-libraries.json with library/discipline classifications.

Python only at this point.

Tested on:

• FabFlee -- Simulation / Uncertainty Quantification
• Topoflow36:
  -- Geospatial Data / Terrain Models -- Signal Processing -- Simulation / Uncertainty Quantification
• Pythia -- Geospatial Data / Terrain Models

[brandomr]

Nice! Any luck finding anything out there that would have a bigger/better mapping? Or do you think we're going to need to expand on this ourselves?

[GoogleSheets]

I think we will need to expand on this ourselves. Searching by discipline name e.g. Network Anaylsis and "python library" usually yields a list; but, I haven't found a classification scheme anywhere.

Implemented some R packages / libraries in latest push. Still need to improve scraping for R libraries/packages, and also for Python (still identifying local imports).