fix: keep bool parameters as bool in sklearn_tuner (#108)

[Nas01010101]

Summary

sklearn_tuner int-casts every discrete numeric hyper-parameter:

if space.paras[k].is_numeric and space.paras[k].is_discrete:
    hyp[k] = int(hyp[k])

BoolPara reports both is_numeric and is_discrete as True, so a boolean hyper-parameter (e.g. a RandomForest's bootstrap) is cast to 0/1. Modern scikit-learn then rejects it:

The 'bootstrap' parameter ... must be an instance of 'bool' ... Got 0 instead.

This is issue #108.

Fix

Special-case BoolPara so boolean parameters stay python bool, and only int-cast the remaining (non-bool) discrete numerics.

Tests

Adds test/test_sklearn_tuner_bool.py — a recorder estimator that asserts the parameter type it receives at fit. Fails before the fix (bool coerced to int), passes after.

1 passed

Closes #108.

[CLAassistant]

All committers have signed the CLA.

[Nas01010101]

Verified this still applies cleanly against current master (0 commits behind) and the added test still fails on revert.

Heads up for merge order: this PR and #123 both touch the for k in hyp: loop body in sklearn_tuner.py (this one changes the cast logic, #123 wraps the loop in warnings.catch_warnings()), so merging both will produce a conflict on whichever lands second. Happy to rebase the other one once you've merged one of them — just ping the branch.