gadget-framework/gadget3

Likelihood distribution aggregating over multiple years / decades

lentinj opened this issue · 2 comments

lentinj commented

From conversation with @vbartolino, @bthe - due to the sparseness of stomach sampling it would be useful to aggregate over more than a single year.

  • Move done_aggregating_f from likelihood_data to g3s_time, so time has a say in when the stock is finished (and it's time to do a comparsion)
  • A g3s_timegroup that can handle year intervals, like we do with age/length, and says we're done aggregating when final step of the end of an interval.
  • Plumb that into likelihood_data, so things like year 1998:2000 are translated appropriately.

See also #29

lentinj commented

@vbartolino we wrote this before I'd started writing g3l_sparsesample, now it exists does that make more sense than multi-year aggregation as above?

vbartolino commented

@vbartolino we wrote this before I'd started writing g3l_sparsesample, now it exists does that make more sense than multi-year aggregation as above?

as far as I've understood (and I may be totally wrong), g3l_sparsesample will help with sparse observations avoiding fitting of artificially generated zeros in the gaps while the specific point here was more related to data which are so sparse that make little sense even at an annual scale but are possibly still informative over multiple-years. I do not think it's priority, but worth keeping it in the big list

  • 👍1