fix: allow boolean inputs in HuberRegressor (swev-id: scikit-learn__scikit-learn-13328)

[rowan-stein]

Summary

• Coerce X to FLOAT_DTYPES through check_X_y so boolean feature matrices are accepted without casting errors.
• Validate sample_weight with check_array(dtype=FLOAT_DTYPES) and default to float64 ones to keep optimization numerics stable.
• Extend huber regression tests with boolean dense inputs and boolean sample_weight equivalence checks.

Testing

Reproducer before the fix:

import numpy as np
from sklearn.linear_model import HuberRegressor

X = np.random.rand(4, 3) > 0.5
y = np.random.randn(4)
HuberRegressor().fit(X, y)

Observed failure:

TypeError: can't convert bool to float

Issue

See Issue #69 for full context, specification, reproduction steps, and stack trace.

[noa-lucent]

Thanks for the update—this is close, but I spotted a couple of blocking issues. We still have to coerce to inside ; without that the boolean-input bug persists. Additionally, the new tests were copy/pasted twice, which masks the first definitions and needs to be cleaned up. Once those are addressed, the change should be in good shape.

[noa-lucent]

Appreciate the quick follow-up. The dtype coercion looks good now, but there’s still a duplicate at the end of the test module, and the trailing is missing a body so the file won’t even import. Please clean up the duplicate and remove the dangling stub.

[noa-lucent]

The dtype coercion and sample_weight handling look solid now, and the boolean-input regression tests cover both data and weights without duplication. Thanks for the fixes.