AMPL 1.8

[paulsonak]

We need to update AMPL for a 1.8 release to accommodate some breaking changes during the GMD containerization process. A few things to work on:

• RDKit version: shouldn't be locked at 2024.3.5
• Remove DGL support for GMD containerization
• Update to latest deepchem & underlying packages

[paulsonak]

• Add dependencies: molvs, umap-learn, imbalanced-learn

[mcloughlin2]

In addition to the items Amanda mentioned, we need to make sure AMPL works with Python 3.10, along with all the package versions that Jeff upgraded to be compatible with Python 3.10. In the testing we (Sarah and I) have done so far, things seem to be generally working, but we've run into some minor problems:

• scikit-learn has changed their file format for random forest models, so those models need to be retrained.
• The xgboost model constructors now have a "device" parameter in place of the "gpu_id" parameter, and there is no n_gpus parameter.

DGL support no longer needs to be removed; Jeff has that working in the container now.

[mcloughlin2]

I added a branch "py310" based on the 1.8.0 branch, specifically for AMPL changes needed to run under Python 3.10 and the updated packages.

[paulsonak]

• we need to update the example model dataset paths to be relative to the "AMPL" top folder for the ones that are shipped with AMPL to allow use across environments

[paulsonak]

• update model compatibility check script