Closed bede closed 2 years ago
I've looked at this again and I can't see an obviously better way of handling this contrived edge case without disabling lazy error handling
Better solution found, I think. Valid sample names are now all coerced to strings by Pandera, whereas these autoincremented indices are floats. I'm now catching these floats in remove_nones_duplicates_empties_ints_from_ld()
and eliminating the sample_name key, meaning that sample-level errors for samples with garbage names are demoted to singleton batch-level errors.
Probably better to ditch the sample_name in these cases, causing the errors to be collapsed into a single error due to redundant errors being pruned? https://oc-collab.gc3.ocs.oraclecloud.com/browse/C900000008-816