I built the PostgresML extension and dashboard from your Github source (v2.7.4 tag) and deployed and configured it on my RHEL8.8 Linux VM environment.

I found that when running the Tutorial 1, step-by-step, using the Dashboard, step 42 failed with the following error:

error returned from database: called `Result::unwrap()` on an `Err` value: PyErr { type: <class 'ModuleNotFoundError'>, value: ModuleNotFoundError("No module named 'sklearn'"), traceback: Some(<traceback object at 0x7fe997260ec0>) }

Caused by:
    called `Result::unwrap()` on an `Err` value: PyErr { type: <class 'ModuleNotFoundError'>, value: ModuleNotFoundError("No module named 'sklearn'"), traceback: Some(<traceback object at 0x7fe997260ec0>) }

From analysis of the Rust source code (https://github.com/postgresml/postgresml/blob/v2.7.4/pgml-extension/src/bindings/sklearn.rs) it seems that some pgml.* functions do not first activate the Python virtual environment, and this seems to be a bug.

For example, the above error does not occur if I prefix the Step 42 query in the notebook cell with:

SELECT pgml.validate_python_dependencies();

(which I know DOES activate the python virtual environment).

Without adding the above workaround, I found that another workaround to this issue was to also install the following in the global python environment:

scikit-learn==1.3.0
xgboost==1.7.6
lightgbm==4.0.0
catboost==1.2

A colleague has since run Tutorial 1 step-by-step using v2.7.8 and reproduced the issue, and I have also looked at the v2.7.8 code, and the issue still seems to be present.
I know that the v2.7.8 dashboard lets you run ALL of the steps in one go, and that DOES work for Tutorial1, but I think that's because the python virtual environment activation happens on a previous step and it maintains the context for subsequent steps because it's done on the same database connection.

Can you confirm this is a bug?