TODO.md

In kidusasfaw/pomp: Statistical Inference for Partially Observed Markov Processes

pomp to-do list

For pomp 2:

• for log-barycentric transformations, check that parameter are contiguous
• do we need basic-component arguments at higher levels of the calling hierarchy for documentation purposes? PROBABLY YES.
• "guide to upgrading to pomp version 2"
  • ~~discussion of stateful objective functions~~
  • ~~include before and after codes for the examples from the JSS paper~~
• more/better examples
  • need examples of objective function methods
• more toy models
• get the metaphors straight! "horses" vs "hitches" vs "workhorses" ("wagons"?)
• manual pages on the organization of the package
  • estimation algorithms
  • diagnostic tools
  • elementary operations: pfilter, probe, simulate, spect
  • workhorses
  • auxiliary functions
  • plotting methods
  • parallelization tools
  • reproducibility tools
  • extending the package
• example for accumvars
• make R-level functions for various distributions and transformations
  • BetaBinom
  • BetaNegBinom
• remove need to specify paramnames when log,logit,barycentric partrans is given? HOW?
• spy methods for derived objects
• better scheme for indicating derivatives and maps in C snippets
• should the default process model be persistence?
  • what would the corresponding dprocess be?

• Kalman filter?

• melt for listies

• coef returns data frame in some circumstances?
• demonstration of Fisher information via pfilter on slice designs?
• new userdata function to get void *?

• ~~simulate with format="d" should return the same thing as simulate %>% as.data.frame~~

• perhaps dinit?

• easier interface for lists of probes in probe
• documentation on mifList, pmcmcList, etc.?
• methods to change data (obs<-)?
  • perhaps recognized data variables, states, covariates in calls to pomp are replaced?
• put Kalman check in 'ou2' test
• MCMC proposals as pomp slots?
• probes as pomp slots?

• what does a generic pomp.fun interface look like?

• should parameter transformations allow renaming of variables?

  • would require attention to rw.sd, e.g.

• weighted particle filter

• in EnKF and EAKF, allow matrices to depend on parameters

For pomp:

• streamline the R snippets so that ... is unnecessary?
  • probably not worth the trouble, since C snippets are so much faster
• add include argument to pomp?
• ~~deprecate obs and states arguments in simulate~~
• graceful stopping for optimizers (at least for nloptr)
• ~~pfilterList object~~
• trap errors for LAPACK
• support for asymmetric MCMC proposals
• one-point SCQL function for possible use in fitting initial conditions
• partial rejection control for pfilter?
• adaptive particle numbers in pfilter (?)
• parallel pfilter algorithm (?)
• plugin for adaptive tau leaping algorithm.
• plugin for compartmental models

Documentation

• FAQ needs update with new issues from @MarieAugerMethe
• ~~Complete rewrite of manual for v. 2~~

Helper packages

• parallel mif-farm and multi-start optimization helpers
• number of processors stored in bake, stew, freeze outputs
• parallel likelihood profiling tool

Examples

• SDE examples
• LPA beetle examples
• budmoth examples

kidusasfaw/pomp documentation built on May 20, 2019, 2:59 p.m.