Andrea-Havron-NOAA
commented
2 months ago prototype in ModularTMBExample, see rcpp_nll, information, and test
Proposed M2 tasks (feedback wanted on naming conventions):
-
[ ] information
- [ ] variable_map: unordered map with unique key (child class name and unique id or unique module id) and pointer. This is set up for all parameters and derived values
- [ ] set_priors
- [ ] loops over all distribution classes with nll_type == "prior", finds correct pointer in variable map and links the distribution's observed_value to the correct parameter using the pointer
- [ ] set_random_effects
- [ ] loops over all distribution classes with nll_type == "re", finds correct pointer in variable map and links the distribution's observed_value to the correct derived value using the pointer
- [ ] set_data
- [ ] loops over all distribution classes with nll_type == "data", finds correct pointer in variable map and links the distribution's expected_value to the correct derived value using the pointer
-
[ ] rcpp_nll: interface for distribution classes
- [ ] string stream: user input used to set key member in distribution
- [ ] set_nll_links: user exposed function that takes required information from user to create unique key to generate string stream
kellijohnson-NOAA
Is your feature request related to a problem? Please describe.
FIMS currently has a flag that allows the user to estimate parameters as random effects, but the model still lacks the infrastructure for implementing RE estimation
Describe the solution you would like.
Generalized infrastructure in the C++ code that allows for hierarchical association of parameters with likelihoods. This infrastructure should allow for both random effect and Bayesian estimation
Describe alternatives you have considered
implement FIMS as penalized likelihood model
Statistical validity, if applicable
State-space stock assessment models have been tested with SAM and WHAM
Describe if this is needed for a management application
No response
Additional context
No response