X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=R%2Fdist.topo.R;h=5777482c35a1555a7abe2865f532e361af7b80d0;hb=40eeb40e48bccc220826860ce0ada4521cbc0148;hp=1eb770af6a12f15948b70388a322eb0f6fcd7992;hpb=6fe5709ee413e5a1a379918a70c64cee05e9ae54;p=ape.git

diff --git a/R/dist.topo.R b/R/dist.topo.R
index 1eb770a..5777482 100644
--- a/R/dist.topo.R
+++ b/R/dist.topo.R
@@ -1,9 +1,9 @@
-## dist.topo.R (2010-01-25)
+## dist.topo.R (2012-03-13)
 
 ##      Topological Distances, Tree Bipartitions,
 ##   Consensus Trees, and Bootstrapping Phylogenies
 
-## Copyright 2005-2010 Emmanuel Paradis
+## Copyright 2005-2012 Emmanuel Paradis
 
 ## This file is part of the R-package `ape'.
 ## See the file ../COPYING for licensing issues.
@@ -11,7 +11,7 @@
 dist.topo <- function(x, y, method = "PH85")
 {
     if (method == "score" && (is.null(x$edge.length) || is.null(y$edge.length)))
-        stop("trees must have branch lengths for Billera et al.'s distance.")
+        stop("trees must have branch lengths for branch score distance.")
     nx <- length(x$tip.label)
     x <- unroot(x)
     y <- unroot(y)
@@ -47,7 +47,6 @@ dist.topo <- function(x, y, method = "PH85")
         for (i in 2:q1) {
             for (j in 2:q2) {
                 if (identical(bp1[[i]], bp2[[j]]) | identical(bp1[[i]], bp2.comp[[j]])) {
-                    if (i == 19) browser()
                     dT <- dT + (x$edge.length[which(x$edge[, 2] == nx + i)] -
                                 y$edge.length[which(y$edge[, 2] == ny + j)])^2
                     found1 <- found2[j] <- TRUE
@@ -72,17 +71,26 @@ dist.topo <- function(x, y, method = "PH85")
     if (any(table(ref) != 1))
         stop("some tip labels are duplicated in tree no. 1")
     n <- length(ref)
-    for (i in 2:length(x)) {
-        if (identical(x[[i]]$tip.label, ref)) next
-        ilab <- match(x[[i]]$tip.label, ref)
-        ## can use tabulate here because 'ilab' contains integers
-        if (any(tabulate(ilab) > 1))
-            stop(paste("some tip labels are duplicated in tree no.", i))
-        if (any(is.na(ilab)))
-            stop(paste("tree no.", i, "has different tip labels"))
-        ie <- match(1:n, x[[i]]$edge[, 2])
-        x[[i]]$edge[ie, 2] <- ilab
-        x[[i]]$tip.label <- NULL
+    Ntree <- length(x)
+    if (Ntree > 1) {
+        for (i in 2:Ntree) {
+            label <- x[[i]]$tip.label
+            if (!identical(label, ref)) {
+                if (length(label) != length(ref))
+                    stop(paste("tree no.", i, "has a different number of tips"))
+                ilab <- match(label, ref)
+                ## can use tabulate here because 'ilab' contains integers
+                if (any(is.na(ilab)))
+                    stop(paste("tree no.", i, "has different tip labels"))
+### <FIXME> the test below does not seem useful anymore
+###            if (any(tabulate(ilab) > 1))
+###                stop(paste("some tip labels are duplicated in tree no.", i))
+### </FIXME>
+                ie <- match(1:n, x[[i]]$edge[, 2])
+                x[[i]]$edge[ie, 2] <- ilab
+            }
+            x[[i]]$tip.label <- NULL
+        }
     }
     x[[1]]$tip.label <- NULL
     attr(x, "TipLabel") <- ref
@@ -95,17 +103,16 @@ prop.part <- function(..., check.labels = TRUE)
     if (length(obj) == 1 && class(obj[[1]]) != "phylo")
         obj <- obj[[1]]
     ## <FIXME>
-    ## class(obj) <- NULL # needed?
+    ## class(obj) <- NULL # needed? apparently not, see below (2010-11-18)
     ## </FIXME>
     ntree <- length(obj)
     if (ntree == 1) check.labels <- FALSE
-    if (check.labels) obj <- .compressTipLabel(obj)
+    if (check.labels) obj <- .compressTipLabel(obj) # fix by Klaus Schliep (2011-02-21)
     for (i in 1:ntree) storage.mode(obj[[i]]$Nnode) <- "integer"
     ## <FIXME>
     ## The 1st must have tip labels
     ## Maybe simply pass the number of tips to the C code??
-    if (!is.null(attr(obj, "TipLabel")))
-        for (i in 1:ntree) obj[[i]]$tip.label <- attr(obj, "TipLabel")
+    obj <- .uncompressTipLabel(obj) # fix a bug (2010-11-18)
     ## </FIXME>
     clades <- .Call("prop_part", obj, ntree, TRUE, PACKAGE = "ape")
     attr(clades, "number") <- attr(clades, "number")[1:length(clades)]
@@ -150,24 +157,31 @@ plot.prop.part <- function(x, barcol = "blue", leftmar = 4, ...)
     mtext(attr(x, "labels"), side = 2, at = 1:n, las = 1)
 }
 
-prop.clades <- function(phy, ..., part = NULL)
+prop.clades <- function(phy, ..., part = NULL, rooted = FALSE)
 {
     if (is.null(part)) {
+        ## <FIXME>
+        ## Are we going to keep the '...' way of passing trees?
         obj <- list(...)
         if (length(obj) == 1 && class(obj[[1]]) != "phylo")
-          obj <- unlist(obj, recursive = FALSE)
+            obj <- unlist(obj, recursive = FALSE)
+        ## </FIXME>
         part <- prop.part(obj, check.labels = TRUE)
     }
-    bp <- .Call("bipartition", phy$edge, length(phy$tip.label),
-                phy$Nnode, PACKAGE = "ape")
-    if (!is.null(attr(part, "labels")))
-      for (i in 1:length(part))
-        part[[i]] <- sort(attr(part, "labels")[part[[i]]])
-    bp <- lapply(bp, function(xx) sort(phy$tip.label[xx]))
+
+    bp <- prop.part(phy)
+    if (!rooted) {
+        bp <- postprocess.prop.part(bp)
+        part <- postprocess.prop.part(part) # fix by Klaus Schliep
+        ## actually the above line in not needed if called from boot.phylo()
+    }
+
     n <- numeric(phy$Nnode)
-    for (i in 1:phy$Nnode) {
-        for (j in 1:length(part)) {
-            if (identical(all.equal(bp[[i]], part[[j]]), TRUE)) {
+    for (i in seq_along(bp)) {
+        for (j in seq_along(part)) {
+            ## we rely on the fact the values returned by prop.part are
+            ## sorted and without attributes, so identical can be used:
+            if (identical(bp[[i]], part[[j]])) {
                 n[i] <- attr(part, "number")[j]
                 done <-  TRUE
                 break
@@ -177,7 +191,8 @@ prop.clades <- function(phy, ..., part = NULL)
     n
 }
 
-boot.phylo <- function(phy, x, FUN, B = 100, block = 1, trees = FALSE)
+boot.phylo <- function(phy, x, FUN, B = 100, block = 1,
+                       trees = FALSE, quiet = FALSE, rooted = FALSE)
 {
     if (is.list(x) && !is.data.frame(x)) {
         if (inherits(x, "DNAbin")) x <- as.matrix(x)
@@ -191,6 +206,8 @@ boot.phylo <- function(phy, x, FUN, B = 100, block = 1, trees = FALSE)
         }
     }
     boot.tree <- vector("list", B)
+    if (!quiet) # suggestion by Alastair Potts
+        progbar <- utils::txtProgressBar(style = 3)
     for (i in 1:B) {
         if (block > 1) {
             y <- seq(block, ncol(x), block)
@@ -198,18 +215,83 @@ boot.phylo <- function(phy, x, FUN, B = 100, block = 1, trees = FALSE)
             boot.samp <- numeric(ncol(x))
             boot.samp[y] <- boot.i
             for (j in 1:(block - 1))
-              boot.samp[y - j] <- boot.i - j
+                boot.samp[y - j] <- boot.i - j
         } else boot.samp <- sample(ncol(x), replace = TRUE)
         boot.tree[[i]] <- FUN(x[, boot.samp])
+        if (!quiet) utils::setTxtProgressBar(progbar, i/B)
     }
+    if (!quiet) close(progbar)
     for (i in 1:B) storage.mode(boot.tree[[i]]$Nnode) <- "integer"
     storage.mode(phy$Nnode) <- "integer"
-    ans <- attr(.Call("prop_part", c(list(phy), boot.tree),
-                      B + 1, FALSE, PACKAGE = "ape"), "number") - 1
-    if (trees) ans <- list(BP = ans, trees = boot.tree)
+
+    pp <- prop.part(boot.tree)
+    if (!rooted) pp <- postprocess.prop.part(pp)
+    ans <- prop.clades(phy, part = pp, rooted = rooted)
+
+    ##ans <- attr(.Call("prop_part", c(list(phy), boot.tree),
+    ##                  B + 1, FALSE, PACKAGE = "ape"), "number") - 1
+    if (trees) {
+        class(boot.tree) <- "multiPhylo"
+        ans <- list(BP = ans, trees = boot.tree)
+    }
     ans
 }
 
+### The next function transforms an object of class "prop.part" so
+### that the vectors which are identical in terms of split are aggregated.
+### For instance if n = 5 tips, 1:2 and 3:5 actually represent the same
+### split though they are different clades. The aggregation is done
+### arbitrarily. The call to ONEwise() insures that all splits include
+### the first tip.
+postprocess.prop.part <- function(x)
+{
+    n <- length(x[[1]])
+    N <- length(x)
+    w <- attr(x, "number")
+
+    drop <- logical(N)
+    V <- numeric(n)
+    for (i in 2:(N - 1)) {
+        if (drop[i]) next
+        A <- x[[i]]
+        for (j in (i + 1):N) {
+            if (drop[j]) next
+            B <- x[[j]]
+            if (length(A) + length(B) != n) next
+            V[] <- 0L
+            V[A] <- 1L
+            V[B] <- 1L
+            if (all(V == 1L)) {
+                drop[j] <- TRUE
+                w[i] <- w[i] + w[j]
+            }
+        }
+    }
+    if (any(drop)) {
+        labels <- attr(x, "labels")
+        x <- x[!drop]
+        w <- w[!drop]
+        attr(x, "number") <- w
+        attr(x, "labels") <- labels
+        class(x) <- "prop.part"
+    }
+    ONEwise(x)
+}
+
+### This function changes an object of class "prop.part" so that they
+### all include the first tip. For instance if n = 5 tips, 3:5 is
+### changed to 1:2.
+ONEwise <- function(x)
+{
+    n <- length(x[[1L]])
+    v <- 1:n
+    for (i in 2:length(x)) {
+        y <- x[[i]]
+        if (y[1] != 1) x[[i]] <- v[-y]
+    }
+    x
+}
+
 consensus <- function(..., p = 1, check.labels = TRUE)
 {
     foo <- function(ic, node) {
@@ -253,6 +335,7 @@ consensus <- function(..., p = 1, check.labels = TRUE)
     ## Get all observed partitions and their frequencies:
     pp <- prop.part(obj, check.labels = FALSE)
     ## Drop the partitions whose frequency is less than 'p':
+    if (p == 0.5) p <- 0.5000001 # avoid incompatible splits
     pp <- pp[attr(pp, "number") >= p * ntree]
     ## Get the order of the remaining partitions by decreasing size:
     ind <- sort(unlist(lapply(pp, length)), decreasing = TRUE,