In some cases, a user will provide common sense constraints e.g. discharge data > admission data. In real data assets, these constraints are often violated (missingness is a big factor here too). Can we turn these semantic clues from constraint violation into features? E.g. a binary feature for each constraint indicating whether it is adhered to in the data or not.
In some cases, a user will provide common sense constraints e.g. discharge data > admission data. In real data assets, these constraints are often violated (missingness is a big factor here too). Can we turn these semantic clues from constraint violation into features? E.g. a binary feature for each constraint indicating whether it is adhered to in the data or not.