Tests passed on my machine. Root cause of the error is TypeError: Cannot cast array data from dtype('int64') to dtype('int32') according to the rule 'safe' at sklearn\utils\sparsefuncs.py:344 when executing test_estimators[VarianceThreshold-check_estimator_sparse_data]. Not sure why that's happening.

@jnothman
Sure. Do you mean setting the variances to np.minimum(np.var(X, 1), np.ptp(X, 1)) or keeping self.variances_ as the variances but using the minimum for comparison against threshold? Also I think the axis should be 0 rather than 1. I'll add a test. There's also the issue of the check failing, which I'm struggling to debug because I can't reproduce it locally.

I mean setting variances_ that way. Unless I'm much mistaken, ptp is a lower bound on various and so will help resolve numerical precision errors

Cool. And the test should be a new function in estimator_checks.py that's yielded by _yield_all_checks if the estimator is a VarianceThreshold?

So a new file for testing this functionality? Most of the existing tests for estimators seem to be in test_common.py as calls to functions in estimator_checks

@jnothman Checks pass, can this now be merged?

Thanks @rlms!