Scoring data with time lap

DocMusher

Hi my RM friends
@IngoMierswa
How to score a model, after deployment, when the data consist of a date which is not the date of scoring.
In my case patient data is consolidated X days after admission, which in other words mean that the data is complete X days after admission. In order to generate a gain vs date chart, not the point in time of the scoring should be used but the admission date of my patient. I can imagine that a variable time lap between the moment that data is generated vs when dataset is complete is a context which is not solely related to health care but also the case in other industries.How to deal with this situation

Telcontar120

If I have understood your question, then I would recommend normalizing all dates as differences relative to the date of admission, which you should be able to do with some Generate Attributes data arithmetic functions (or if you have the data in time series format using the time series operators).  This will give you generalized dates in the form {number of days since admission} rather than specific dates. You'll want to rerun your original models to ingest this version of the time data, which will then allow you to generalize patterns and your models won't get hung up when they are fed data from a later time period (as long as the timeframe of the analysis is pretty consistent with respect to the date of admission).

Telcontar120

If I have understood your question, then I would recommend normalizing all dates as differences relative to the date of admission, which you should be able to do with some Generate Attributes data arithmetic functions (or if you have the data in time series format using the time series operators).  This will give you generalized dates in the form {number of days since admission} rather than specific dates. You'll want to rerun your original models to ingest this version of the time data, which will then allow you to generalize patterns and your models won't get hung up when they are fed data from a later time period (as long as the timeframe of the analysis is pretty consistent with respect to the date of admission).