sandbox/TODO.md

In queezzz/qzmle: Maximum likelihood estimation

to do (in no particular order)

• make a qzmle repository that contains the skeleton of a qzmle R package that will eventually be a cleaner version (replacement? refactoring tool?) for bbmle

  • devtools, usethis
  • copy mkfun from sandbox/tmp.R: think about a good name for it
  • think about conventions: how are you going name functions? what code style are you going to use? https://style.tidyverse.org/ (styler package?)

• extend mkfun to deal with more parameters

  • what if PDFs depend on multiple parameters?
  • what if PDF parameters depend on multiple model parameters?
  • for a PDF with multiple parameters:
    • write out the chain-rule process by hand and see how it generalizes to >1 parameter
    • for 1-parameter: dL/d(a,b,c) = dL/d(lambda) * d(lambda)/d(a,b,c) = ({\partial{lambda}/\partial{a}, \partial{lambda}/\partial{b}, \partial{lambda}/\partial{c}})
    • {mean, sd} {a, b, c}: L <- {mean, sd}, mean -> {a,b} sd -> {c}

Which parts of a formula are parameters? Which parts are data?

we have p PDF parameters (e.g. for Poisson, p=1 {lambda}) and for each parameter we have p_i model parameters (e.g. b0+b1*latitude^2, p_1=2)

we have N observations, so dL/dlambda = vector of length N d lambda/d(b0) = vector of length N d lambda/d(b1) = vector of length N

queezzz/qzmle documentation built on Nov. 24, 2021, 6:34 p.m.