V2: Fix small hpopt bugs and add example notebook

[hwpang]

Description

Add an example hpopt notebook. Happy to take suggestion on the notebook.
Found some small bugs related to input type and saving hpopt results while working on the notebook. Fixed those.

Example / Current workflow

n/a

Bugfix / Desired workflow

n/a

Questions

n/a

Relevant issues

n/a

Checklist

• linted with flake8?
• (if appropriate) unit tests added?

[shihchengli]

Thanks for making this PR. There are two minor things you might consider adding to the notebook:

1. Include metrics within the model (e.g., metric_list = [nn.metrics.RMSEMetric(), nn.metrics.MAEMetric()]) in the notebook.
2. Add the ASHAScheduler to the notebook, as we use it by default.

[shihchengli]

The test fails because the filename has changed. Changing the filename of "best_params.json" to "best_config.json" in test_hyperopt_quick within tests/unit/utils/test_save_load_mol.py can fix it.

[kevingreenman]

should we add another notebook with an example of using hpopt with optuna, in addition to the hyperopt example? (after #783 is resolved) see #774

[shihchengli]

@hwpang here are my final comments about the notebook

1. replace the uniform with qrandint in the search space
2. Since the default scheduler is changed to FIFO in Change the default trial scheduler to FIFOScheduler #896, please replace the ASHAScheduler with FIFOScheduler
3. As the optuna is supported via Add Optuna support for HPO #888, instead of giving a new notebook about how to use optuna, you can add a comment in the file before the HyperOptSearch is used to generate the object of the search algorithm in the example, indicating that they can also use optuna by search_alg = OptunaSearch(), and change the notebook file name to hpopting.ipynb
4. remove the last empty cell
5. (Optional) you can consider adding ray.init() before running ray

[hwpang]

@shihchengli Thanks for reviewing! I have updated according to your comments

[shihchengli]

There is a RayTaskError(OutOfMemoryError) in the notebook. I guess you are using estes cluster to test this notebook, and those two GPUs in the cluster are fully occupied now. You can try running this on the other cluster or use a CPU to test it, as we will also run the test in CI and we only have CPUs. I think it is fine to use the CPU in the notebook and point out the way to use the GPU.

[shihchengli]

LGTM

[shihchengli]

I noticed a minor thing that needed to be changed in the notebook. You can remove find_unused_parameters=True from the notebook as we fixed it via #883.

[shihchengli]

The error can be solved by changing the CI configuration to run test on hpopting.ipynb exclusively with Python 3.11.

[shihchengli]

You can do something like this, although it's not very elegant.

  - name: Test notebooks (excluding hpopting.ipynb)
    shell: bash -l {0}
    run: |
      python -m pip install matplotlib
      pytest --no-cov -v --nbmake $(find examples -name '*.ipynb' ! -name 'hpopting.ipynb')
  - name: Test hpopting.ipynb (Python 3.11 only)
    if: ${{ matrix.python-version == '3.11' }}
    shell: bash -l {0}
    run: |
      python -m pip install matplotlib
      pytest --no-cov -v --nbmake examples/hpopting.ipynb

[hwpang]

@shihchengli Thanks for the review and the suggestion! I have updated the CI accordingly