What is the bootstrap round ('num_rounds' param in mlxtend.evaluate bias_variance_decomp)?

[diman82]

Describe the documentation issue

Suggest a potential improvement or addition

[rasbt]

Hi there. It's the number of bootstrap rounds. I.e., how many bootstrap samples are being drawn. Let me know if you need additional clarification.

[diman82]

Hi.
Bootstrap samples are being drawn from where? From viewing the source code it seems, that the function randomly re-samples data (from the train set) and runs num_rounds of iterations (.fit()) with the newly sampled data.

[rasbt]

Yes, that correct. Thanks for bringing this up, this should be described in the docstring; will fix that!