hyperparameter search

[dannyoo]

Greetings,

What would be the best way to perform a hyperparameter search on the classifier?

Thanks

[mirand863]

Hi @dannyoo,

There are different ways to perform hyperparameter search.

Scikit-learn is probably the easiest to use, with approaches such as grid search and randomized optimization https://scikit-learn.org/stable/modules/grid_search.html

For HiClass' paper, I chose to use Hydra and its optuna sweeper https://hydra.cc/docs/intro/ due to automatic submission in a slurm cluster and other more advanced search algorithms that I wanted to test. The learning curve is a bit more steep, but it has more features. In the end, you have to evaluate which one is more appropriate for your use case.

It is also important to read the documentation for the local classifier and the local hierarchical classifier to identify any hyperparameters that could be interesting to tune. For example, for logistic regression you can try different solvers and penalties, while for the local classifier per node you could test different policies.

I hope this answers your question.

Best regards,
Fabio