Error in chol.default(R) :the leading minor of order 1 is not positive definite as using the classif.naiveBayes

[Seager1989]

Error in chol.default(R) :the leading minor of order 1 is not positive definite as using the classif.naiveBayes

[pat-s]

possibly related #2738

[Seager1989]

Thank you for your response. I understand this may be not the problem of mlr, but why the configureMlr can not handle this type of error by setting on.learner.error = "warn"?

[Seager1989]

I checked the error again and found it may be related to the Kriging model fitting after the evaluation of the initial DOE. This may be related to the mlr.

This is because the error appeared after four evaluations even using different datasets. I have only one hyperparameter for tunning, so the default initial size of DOE is 4. The error is due to the fitting of the Kriging surrogate model in the sequential model-based optimization. This is why the error always appears after four evaluations. I do not know if you have some ideas to change something to avoid this type of error.

[pat-s]

If you think this an error with MBO please open an error including a reprex in mlr-org/mlrMBO.

[Seager1989]

Thank you for your suggestion. I am also a little bit confused about the boundary of mlr and mlrMBO. I did use the makeTuneControlMBO and tuneParams, which are the function of mlr. In this way, this is an issue on the mlr or mlrMBO?

[pat-s]

{mlrMBO} provides tuning methods for {mlr}.

The error looks like a learner issue so it relates to {mlr}.
Note that {mlr} is officially deprecated - we suggest to use {mlr3} instead.

For more help, please post a full reproducible example.

[pat-s]

Closing due to missing reprex and no response.