Add European inverse Pareto distribution (invpareto_eur) - #170
Draft
joethorley wants to merge 10 commits into
Draft
joethorley wants to merge 10 commits into
joethorley wants to merge 10 commits into
Conversation
Add the European form of the inverse Pareto distribution as a new user-selectable distribution `invpareto_eur`, and clarify that the existing `invpareto` is the North American form. The European version is unbounded above and uses the same shape/scale parameterisation as `actuar::dinvpareto()`, making it a more realistic stand-alone SSD than the bounded North American form. Its scale is a free parameter estimated by maximum likelihood (no boundary fixing). - Add TMB log-likelihood `ll_invpareto_eur` and register it. - Add `ssd_pinvpareto_eur()`, `ssd_qinvpareto_eur()`, `ssd_rinvpareto_eur()`, `ssd_einvpareto_eur()` and internal helpers, mirroring `invpareto`. - Register `invpareto_eur` in `dist_data` (bcanz = FALSE, valid = FALSE) so it is fitted in isolation like `invpareto`. - Document the American vs European distinction in the distributions vignette and parameter docs. - Add tests for both forms, including an independent check that the fitted mle and the p/q functions match the actuar parameterisation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Mark `invpareto` and `invpareto_eur` as valid = TRUE in `dist_data` so that both can be fitted alongside other distributions and enrolled in model averaging, rather than only in isolation. - Set valid = TRUE for both forms in `dist_data`. - Add invpareto.* and invpareto_eur.* parameters to `ssd_pmulti()`, `ssd_qmulti()` and `ssd_rmulti()` so the model-averaging machinery accepts them. - Update the distributions vignette: both forms can now be model-averaged, with a caution that the North American `invpareto` is bounded above so its contribution to upper-tail estimates should be interpreted with care. - Update tests and snapshots: both forms now appear in `ssd_dists_all()`, `invpareto_eur` joins `ssd_dists_shiny()`, `ssd_emulti()` gains both forms, and the model-averaged cdf plots now include both. Replace the invpareto_eur isolation test with one confirming it can be fitted alongside other distributions. Both forms remain excluded from the default (bcanz) candidate set and receive zero weight in the default model-averaging skeleton, so default results are unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fix the remaining test failures surfaced by R CMD check after marking both inverse Pareto forms valid: - Rewrite the `lnorm + invpareto` isolation test to confirm the combination now fits (the isolation rule no longer applies since there are no invalid-likelihood distributions). - Update the `estimates(all_estimates = TRUE)` snapshot to include the new weight-0 invpareto and invpareto_eur entries. - Update the `tidy_unstable_anon_e` snapshot now that `ssd_dists(bcanz = FALSE)` includes both invpareto forms. Full R CMD check passes with 0 errors, 0 warnings, 0 notes. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Both inverse Pareto forms are bounded/limiting distributions whose likelihoods are not comparable to the standard candidates for AICc model averaging, so they must be fitted in isolation (valid=FALSE), matching the long-standing treatment of the North American invpareto on dev. - Revert invpareto valid TRUE->FALSE (the PR had inadvertently flipped it). - Set invpareto_eur valid=FALSE. - Update tests: invpareto_eur must be fitted in isolation; drop it from ssd_dists_shiny(); remove the added invpareto-with-others test. - Regenerate dist_data and the valid-filtered snapshots. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ructural assertions The hc/hp bootstrap confidence-interval snapshots (hc_boron, hp_boron) are not reproducible across platforms (bootstrap refits diverge with different BLAS/LAPACK), so they failed on Linux/Windows CI. Assert structural properties (finite, ordered confidence limits) instead. The exact snapshots are preserved in the ssdtests package (poissonconsulting/ssdtests@add-invpareto-eur). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
# Conflicts: # NAMESPACE
`(q / (q + scale))^shape` is only a CDF for q > 0. For negative q it is negative, or positive for an even shape, so `ssd_pinvpareto_eur(-2)` returned 8. Return exactly 0 on and below the lower support limit, the same contract #195 established for the bounded inverse Pareto. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The European inverse Pareto is a regular two-parameter distribution with an interior maximum likelihood estimate, unlike the bounded North American form whose scale sits on the data maximum, so there is no reason to restrict it to being fitted in isolation. Register it with `valid = TRUE`, which the distributions article already claimed. It therefore joins `ssd_dists_all()` and `ssd_dists_shiny()`, the `ssd_emulti()` defaults, and the fits-of-everything snapshots. The `ssd_match_moments()` snapshot also shifts for the existing distributions because adding a distribution changes the order in which random draws are consumed under the fixed seed, as it did for `ltriangle`. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds the European form of the inverse Pareto distribution as a new user-selectable distribution
invpareto_eur, and makes explicit that the existinginvparetois the North American form, following the convention the distributions article already uses (European = unbounded, as inactuar; North American = bounded, as inextraDistr).The European version is unbounded above and uses the same
shape/scaleparameterisation asactuar::dinvpareto(). Both parameters are free MLE parameters with an interior optimum, so unlike the bounded North American form it is registered withvalid = TRUEand can be model averaged alongside the other distributions. It is not in the BCANZ default set.Changes
src/TMB/ll_invpareto_eur.hpp: negative log-likelihood, log-densitylog(shape) + log(scale) + (shape - 1) log y - (shape + 1) log(y + scale), censored branch via the closed-form CDF.R/invpareto_eur.R:ssd_pinvpareto_eur(),ssd_qinvpareto_eur(),ssd_rinvpareto_eur(),ssd_einvpareto_eur(), starting values (shape = 1,scale= median), and thep/qbackends. The CDF is exactly 0 for non-positive concentrations, the contract qdist() ignores bounded support at the extreme quantiles #195 established forinvpareto.dist_data:invpareto_eurwithbcanz = FALSE,tails = TRUE,npars = 2,valid = TRUE,bound = FALSE. It therefore appears inssd_dists_all()andssd_dists_shiny().ssd_pmulti()/ssd_qmulti()/ssd_rmulti()gaininvpareto_eur.*arguments;params.Rdocuments them.test-invpareto_eur.Rcoverstest_dist(), agreement withactuarforp/qand the MLE, the anon_a and boron fits,ssd_hc()/ssd_hp()confidence limits, the clamp on non-positive concentrations, and model averaging withlnorm. Snapshots that enumerate all distributions are updated.Verification
Merged with current
dev.ssd_pinvpareto_eur()matchesactuar::pinvpareto()exactly and the quantile to 1.6e-11. Fits are scale invariant: multiplyingccme_boronby 1000 leavesshapeunchanged to seven digits and scalesscaleby exactly 1000. Endpoints follow #195:ssd_qinvpareto_eur(c(0, 1))is0, Infandssd_hc(fit, proportion = c(0, 1))returns the same.Full suite 1477 passing, 0 failures;
R CMD check0 errors, 0 warnings, 0 notes;air format --checkclean.Notes
The
ssd_match_moments()snapshot changes for the existing distributions as well as gaining aninvpareto_eurrow: adding a distribution changes the order in which random draws are consumed under the fixed seed, as it did forltriangle. Becausetails = TRUEandvalid = TRUE,ssd_dists_shiny()now listsinvpareto_eur, so shinyssdtools will offer it once it picks up this release.