If a method's results are not exactlyduplicated, then numerical difficulties still arise (e.g., SL.glm with intercept-only model may provide a mean estimate to a different level of precision than SL.mean though duplicated does not pick up on this).
This pull request adds a tolerance level by rounding the columns of Z before checking for duplicates. I've set the default to 8 digits, which seems like a reasonable level, though further testing may reveal a more/less stringent threshold will suffice.
If a method's results are not exactly
duplicated, then numerical difficulties still arise (e.g.,SL.glmwith intercept-only model may provide a mean estimate to a different level of precision thanSL.meanthoughduplicateddoes not pick up on this).This pull request adds a tolerance level by rounding the columns of
Zbefore checking for duplicates. I've set the default to 8 digits, which seems like a reasonable level, though further testing may reveal a more/less stringent threshold will suffice.