Optimise GM99 multi-resident establishment (issue #34)#35
Merged
Conversation
Two changes to src/GM99.cpp: 1. Tighter Poisson truncation in estab(): reduce Kmax from N + 10*sqrt(N+1) + 20 to N + 6*sqrt(N+1) + 6. The tail beyond the new cutoff is < 1e-10, so accuracy is unchanged but each dimension of the Cartesian odometer has ~2x fewer terms (e.g. N=1: 31 -> 16). 2. Replace the fixed-point iteration N_i <- N_i * W_i (up to 10000 steps, absolute tolerance) in gm99_equilibrium() with Newton's method on log W_i(N) = 0 in log-N space, using a numerical Jacobian computed by forward differences. Newton converges quadratically so only ~10-30 steps are needed vs thousands for the old fixed-point, which could oscillate slowly for multi-resident systems. Competitive exclusion is handled by removing near-zero residents (N < 1e-8, log W < 0) from the active Jacobian system. Before/after benchmarks (alpha=7, beta=15, R=1): 2-resident equilibrium: ~3.5 ms -> 0.11 ms (>30x) 3-resident equilibrium: ~1100 ms -> 4.7 ms (>230x, target was <100 ms) assembler step at 3 residents: ~6000 ms -> <0.1 ms (>60000x) All 218 tests pass unchanged. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adapt the plant-family build conventions to regnans: - Makefile: adapted from plant/Makefile, dropping RcppR6/vignettes/ website targets (regnans uses plain Rcpp::compileAttributes and has no vignettes or pkgdown). Targets: compile, attributes, roxygen, test, install, build, check, clean. - .github/workflows/R-CMD-check.yaml: adapted from plant-dev2's modern r-lib/actions workflow (plant itself only has legacy Travis/AppVeyor). Runs R CMD check on Windows + macOS. - DESCRIPTION: add Remotes (traitecoevo/plant, smbache/loggr) so CI can install the GitHub-only dependencies. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
dfalster
added a commit
that referenced
this pull request
Jun 29, 2026
Brings `R CMD check` to **Status: OK** (was 6 WARNINGs + 3 NOTEs, plus
an install failure and a CI vignette-build error on macOS).
### Fixes
- **Install failure** — `man/harness_gm99.Rd` reworded so R 4.6's
`parse_Rd` no longer reads apostrophes (`x'`, `paper's`, `Daniel's`) as
quote delimiters (was: *"unexpected end of input"*).
- **Latent runtime bug** — dropped a stray `ctrl = ctrl` arg in
`demography_solve_equilibrium_solve()` (unused-argument byte-compile
note).
- **NAMESPACE / dependencies** — added `@importFrom` for
stats/utils/ggplot2/dplyr/tibble/purrr + `utils::globalVariables()` for
NSE columns; moved `dplyr`/`tidyr`/`mlr3learners` from Depends to
Imports; declared
`bbotk`/`data.table`/`loggr`/`mlr3`/`mlr3mbo`/`paradox`; dropped unused
`progress`; `requireNamespace(mlr3learners)` instead of `require()`.
- **Documentation** — synced `@param` blocks with signatures
(`community_start`, `community_solve_singularity_1D`,
`community_fitness_landscape`); described empty `@param ...`; made
`community_fitness_landscape_bayesopt` internal; fixed broken
`\link{assembler_stochastic_naive}`.
- **Stale generated files** — regenerated `RcppExports.R` +
`gm99_equilibrium.Rd` to match the committed Newton `GM99.cpp`
(`200L`/`1e-10`). Master currently still has the old `10000L`/`1e-12` in
these generated files, so this also clears the codoc mismatch now
present on master.
- **DESCRIPTION** — complete-sentence `Description`; removed `LazyData`
(no `data/`).
- **.Rbuildignore** — `.claude`, `.github`, `.plant-interface-version`,
`AGENTS.md`, dev `Makefile` (GNU-extensions warning).
### Vignettes → staging
The three WIP vignettes (`assembly_fitmax`, `assembly_stochastic`,
`solving_attractors`) ran the full plant SCM on every build (the source
of the macOS CI `creating vignettes ... ERROR`). Moved out to
`overstorey_staging/` as `.qmd`.
### Not touched
`loggr` is left as-is (Imports + `Remotes: smbache/loggr`), matching
`plant`, which uses it and passes on Windows. The earlier Windows
`pkgdepends` resolution error on #35 looks like a transient/knock-on
solve failure rather than a real loggr problem.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
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.
Summary
Fixes the exponential-in-species-count slowdown in
gm99_equilibrium()reported in #34. Two changes tosrc/GM99.cpp:Tighter Poisson truncation in
estab(): Kmax goes fromN + 10·√(N+1) + 20toN + 6·√(N+1) + 6. Tail probability beyond the new cutoff is < 1e-10, accuracy is unchanged, but each dimension of the Cartesian odometer has ~2× fewer terms (e.g. N=1: 31 → 16 terms).Newton's method replaces the fixed-point iteration
N_i ← N_i · W_iingm99_equilibrium()for the multi-resident case. Newton solveslog W_i(N) = 0in log-N space with a numerical Jacobian (forward differences). It converges quadratically so only ~10–30 steps are needed vs. thousands for the old fixed-point, which could oscillate slowly near equilibrium. Competitive exclusion is handled by removing residents with near-zero density and negative log fitness from the active Jacobian subsystem.Before/after benchmarks (alpha=7, beta=15, R=1)
The 3-resident target from the issue (< 100 ms) is met with room to spare (4.7 ms).
Test plan
test-harness-gm99.R: all oracle tests pass (W_m(m)=1, closed-form g, CSS/branching at αR=4.5/7, αR·βR scaling)test-assembly.R: all assembly tests passCloses #34.
🤖 Generated with Claude Code