doc/interpretation.md

In richelbilderbeek/pirouette: Create a Bayesian Posterior from a Phylogeny

Interpretation

Relevance of comparing error distributions

Compare with | TB | WG | WB -----------------|--------|--------|-------- TG | Y | Y | N TB | . | N | Y WG | . | . | Y WB | . | . | .

Legend:

• TG: true tree, generative model
• TB: true tree, best candidate model
• WG: twin tree, generative model
• WB: twin tree, best candidate model

Inpretation of comparing relevant error distributions

Condition | Expectation | Interpretation --------------|-----------------|------------------------------------------------------------------------------------------------------- $TG > TB$ | Unexpected | Novel tree prior is more related to best candidate than hand-picked tree prior $TG ≈ TB$ | Possible | Hand-picked tree prior is just as suitable as the best candidate tree prior $TG < TB$ | Expected | Hand-picked tree prior is the most related tree prior $TG > WG$ | Unknown | Novel tree prior important $TG ≈ WG$ | Unknown | Novel tree prior unimportant $TG < WG$ | Unexpected | Twinning procedure increases inference errors when using hand-picked tree prior $TB > WB$ | Expected | Impact of novel tree prior cannot be compensated for by model selection: twin tree with low likelihood? $TB ≈ WB$ | Possible | Best candidate tree priors perform equally well in true and twin tree: true and twin tree similar? $TB < WB$ | Unexpected | Twinning procedure increases inference errors when using best tree prior candidate $WG > WB$ | Unexpected | Hand-picked tree prior (that equals the twinning tree prior!) worse than best candidate tree prior $WG ≈ WB$ | Possible | Twin tree fits equally well to the hand-picked and best candidate tree prior $WG < WB$ | Expected | Hand-pick tree prior (that equals the twinning tree prior) performs as expected

This interpretation assumes that the operators ($<$, $\approx$ and $>$) to compare distributions are defined.

Legend:

• TG: true tree, generative model
• TB: true tree, best candidate model
• WG: true tree, generative model
• WB: true tree, best candidate model

richelbilderbeek/pirouette documentation built on Oct. 18, 2023, 3:09 p.m.