Bound Rogue SPR candidates with parent dual prices

.AGENTS/memory/Optimizations.md

@@ -978,3 +978,12 @@ in TreeDist — every sort is `std::sort` / `std::stable_sort`.
   of those end below the cutoff (candidate rejected); the row bound exits
   ~2%, a column-max bound would add ~1%.  Further gains need a tighter
   MCI upper bound or LAP warm starts between candidates.
+
+## Rogue SPR: parent dual prices (PR #215, 2026-10-14)
+Full write-up: `dev/rogue_mci_bound_findings.md`.  Parent column duals bound
+82–88% of wasted LAPs; remaining LAPs warm-start (2–4×).  −51% at 80 tips,
+−65% at 150.  Size-profile / compatibility bounds ~100× too loose; progressive
+row-filling skips ≤1.5% of fill.  Next lever: build child cost matrices
+incrementally from the parent (97.6% of candidates drop one leaf); the fill is
+now 37–55% of runtime.  Parent prices are computed lazily (first candidate
+past the row bound).

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: TreeDist
 Title: Calculate and Map Distances Between Phylogenetic Trees
-Version: 2.15.0.9006
+Version: 2.15.0.9007
 Authors@R: c(person("Martin R.", "Smith",
                     email = "martin.smith@durham.ac.uk", 
                     role = c("aut", "cre", "cph", "prg"), 

NEWS.md

@@ -1,4 +1,4 @@
-# TreeDist 2.15.0.9006 (development)
+# TreeDist 2.15.0.9007 (development)
 
 ## Bug fixes
 
@@ -8,6 +8,11 @@
 
 ## Performance
 
+- `SPRDist(method = "Rogue")` bounds each candidate's clustering information
+  with the optimal dual prices of the parent pair, skipping most assignment
+  problems and warm-starting the rest: c. 50% faster at 80 leaves and 65% at
+  150 leaves on unrelated trees.
+
 - Improve performance when comparing many similar trees with
   `ClusteringInfoDistance()`, `MatchingSplitDistance()`, `InfoRobinsonFoulds()`
   and `JaccardRobinsonFoulds()`.

codemeta.json

@@ -8,7 +8,7 @@
   "codeRepository": "https://github.com/ms609/TreeDist/",
   "issueTracker": "https://github.com/ms609/TreeDist/issues/",
   "license": "https://spdx.org/licenses/GPL-3.0",
-  "version": "2.15.0.9006",
+  "version": "2.15.0.9007",
   "programmingLanguage": {
     "@type": "ComputerLanguage",
     "name": "R",
@@ -571,7 +571,7 @@
           "email": "martin.smith@durham.ac.uk"
         }
       ],
-      "name": "TreeDist: Distances between Phylogenetic Trees. R package version 2.15.0.9006",
+      "name": "TreeDist: Distances between Phylogenetic Trees. R package version 2.15.0.9007",
       "identifier": "10.32614/CRAN.package.TreeDist",
       "@id": "https://doi.org/10.32614/CRAN.package.TreeDist",
       "sameAs": "https://doi.org/10.32614/CRAN.package.TreeDist",

dev/rogue_bound/add_mid.R

@@ -0,0 +1,10 @@
+suppressMessages(library(TreeTools, lib.loc = "/root/Rlib"))
+src <- readLines("make_workload.R")
+end <- grep("^set.seed", src)[1] - 1
+eval(parse(text = src[1:end]))
+wl <- readRDS("workloads.rds")
+set.seed(777)
+wl$mid80 <- lapply(1:12, function(i) NearPair(80, 15))
+wl$mid150 <- lapply(1:6, function(i) NearPair(150, 30))
+saveRDS(wl, "workloads.rds")
+cat(names(wl), "\n")

dev/rogue_bound/alt_summary.R

@@ -0,0 +1,17 @@
+dir <- "/tmp/claude-0/-home-claude/37427f5a-6383-50c4-a6e2-4664447d3c94/scratchpad/bound/csv2"
+out <- NULL
+for (f in c("rand30", "rand30d2", "rand80", "rand150", "near80", "near150")) {
+  x <- read.csv(file.path(dir, paste0(f, ".csv")))
+  lapCalls <- x$ub_row >= x$min_score
+  w <- x[lapCalls & x$lap_exact < x$min_score, ]
+  Rej <- function(b) { v <- w[[b]]; v[is.na(v)] <- Inf; 100 * mean(v < w$min_score) }
+  out <- rbind(out, data.frame(
+    wl = f, n_wasted = nrow(w),
+    cr = Rej("ub_cr"), rc = Rej("ub_rc"), best_cr_rc = Rej("ub_alt1"),
+    alt3 = Rej("ub_alt3"), alt10 = Rej("ub_alt10"),
+    pv = Rej("ub_pv_minv"), pv_max = Rej("ub_pv_maxv"), pv_mean = Rej("ub_pv_mean"),
+    pu = Rej("ub_pu"), pv1 = Rej("ub_pv_alt1"), pv3 = Rej("ub_pv_alt3"),
+    pvarr = Rej("ub_pv_warmarr"), prev = Rej("ub_prev"),
+    arr = Rej("ub_arr"), rt = Rej("ub_rt")))
+}
+print(format(out, digits = 3), row.names = FALSE)