version 4.1

Author: Marco Scutari

Changelog

@@ -1,3 +1,33 @@
+bnlearn (4.1)
+
+  * fixed memory corruption in dsep() (thanks Dominik Muller).
+  * added the marginal uniform prior.
+  * fixed the optimized score cache for the Castelo & Siebes and for the
+     marginal uniform priors, which were affected by several subtle bugs.
+  * bn.cv() now implements a "custom-folds" method that allows to manually
+     specify which observation belongs to each fold, and folds are not
+     constrained to have the same size.
+  * fixed checks in the C code involving R objects' classes; they failed
+     when additional, optional classes were present (thanks Claudia Vitolo).
+  * fixed cpdag() handling of illegal arcs that are part of shielded
+     colliders (thanks Vladimir Manewitsch).
+  * removed misleading warning about conflicting v-structures from cpdag().
+  * rsmax2() and mmhc() now return whitelists and blacklists as they are
+     at the beginning restrict phase (thanks Vladimir Manewitsch).
+  * bn.fit() can now fit local distributions in parallel, and has been mostly
+     reimplemented in C for speed (thanks Claudia Vitolo).
+  * added an impute() function to impute missing values from a bn.fit object.
+  * fixed loss functions for data in which observations have to be dropped
+     for various nodes (thanks Manuel Gomez Olmedo).
+  * added an all.equal() method to compare bn.fit objects.
+  * added a "by.node" argument to score() for decomposable scores (thanks
+     Behjati Shahab).
+  * added warning about partially direct graphs in choose.direction() and
+     improved its debugging output (thanks Wei Kong).
+  * added spouses(), ancestors() and descendats().
+  * fixed a segfault in predict(..., method = "lw") with discrete BNS and
+     sparse CPTs that included NaNs.
+
 bnlearn (4.0)
 
   * fixed memory usage in aracne(), chow.liu() and tree.bayes() (thanks

DESCRIPTION

@@ -2,8 +2,8 @@ Package: bnlearn
 Type: Package
 Title: Bayesian Network Structure Learning, Parameter Learning and
         Inference
-Version: 4.0
-Date: 2016-05-12
+Version: 4.1
+Date: 2017-02-09
 Depends: R (>= 2.14.0), methods
 Suggests: parallel, graph, Rgraphviz, lattice, gRain
 Author: Marco Scutari
@@ -22,11 +22,11 @@ Description: Bayesian network structure learning, parameter learning and
   included, as well as support for parameter estimation (maximum likelihood
   and Bayesian) and inference, conditional probability queries and
   cross-validation. Development snapshots with the latest bugfixes are
-  available from www.bnlearn.com.
+  available from <http://www.bnlearn.com>.
 URL: http://www.bnlearn.com/
 License: GPL (>= 2)
 LazyData: yes
 NeedsCompilation: yes
-Packaged: 2016-05-15 21:32:16 UTC; fizban
+Packaged: 2017-02-09 16:21:05 UTC; fizban
 Repository: CRAN
-Date/Publication: 2016-05-16 14:47:13
+Date/Publication: 2017-02-09 18:20:13

MD5

@@ -34,6 +34,9 @@ export(
   "parents<-",
   "children",
   "children<-",
+  "spouses",
+  "ancestors",
+  "descendants",
   "root.nodes",
   "leaf.nodes",
   "nnodes",
@@ -101,6 +104,7 @@ export(
   "dsep",
   "discretize",
   "dedup",
+  "impute",
   "configs",
   "relevant",
   "read.bif",
@@ -128,6 +132,7 @@ importFrom("graphics", "abline", "arrows", "lines", "plot", "points",
 S3method(as.bn, fit)
 
 S3method(all.equal, bn)
+S3method(all.equal, bn.fit)
 S3method(print, bn)
 S3method(plot, bn)
 S3method(AIC, bn)

NAMESPACE

@@ -88,7 +88,7 @@ arc.strength.score = function(network, data, score, extra, debug = FALSE) {
   reference.score = per.node.score(network = network, score = score,
                       targets = nodes, extra.args = extra, data = data)
   # check whether the score is decomposable.
-  decomp = is.score.decomposable(score, nodes, extra)
+  decomp = is.score.decomposable(score, extra)
 
   if (debug) {
 

R/arc.strength.R

@@ -274,7 +274,7 @@ vstruct.apply = function(arcs, vs, nodes, strict, debug = FALSE) {
     temp = set.arc.direction(v["y"], v["x"], arcs)
     temp = set.arc.direction(v["z"], v["x"], temp)
 
-    if(!is.acyclic(temp, nodes, directed = TRUE)) {
+    if (!is.acyclic(temp, nodes, directed = TRUE)) {
 
       if (debug) {
 

R/backend-indep.R

@@ -160,6 +160,10 @@ choose.direction.test = function(x, arc, data, test, alpha, B, debug = FALSE) {
   # because if they, too, are parents of the node to be tested
   # they _do_ belong there; if they are not, the node distribution
   # does not depend on them so they are largely irrelevant.
+  if (any(arc %in% undirected.arcs(x)))
+    warning("the graph is not completely directed around ", arc[1], " and ",
+      arc[2], ", treating nodes connected by undirected arcs as parents.")
+
   choose.direction.test.pvalue = function(arc) {
 
     if (!is.legal.arc(arc, test, data)) {
@@ -208,6 +212,9 @@ choose.direction.score = function(x, data, arc, score, extra.args, debug = FALSE
   # because if they, too, are parents of the node to be tested
   # they _do_ belong there; if they are not, the node distribution
   # does not depend on them so they are largely irrelevant.
+  if (any(arc %in% undirected.arcs(x)))
+    warning("the graph is not completely directed around ", arc[1], " and ",
+      arc[2], ", treating nodes connected by undirected arcs as parents.")
 
   x$nodes[[arc[1]]]$parents = parents.backend(x$arcs, arc[1], undirected = TRUE)
   x$nodes[[arc[2]]]$parents = parents.backend(x$arcs, arc[2], undirected = TRUE)
@@ -218,7 +225,7 @@ choose.direction.score = function(x, data, arc, score, extra.args, debug = FALSE
   reference.score = per.node.score(network = x, score = score,
                       targets = arc, extra.args = extra.args, data = data)
   # check whether the score is decomposable.
-  decomp = is.score.decomposable(score, names(x$nodes), extra.args)
+  decomp = is.score.decomposable(score, extra.args)
 
   # compare the scores of the two networks.
   choose.direction.score.delta = function(arc) {
@@ -230,7 +237,7 @@ choose.direction.score = function(x, data, arc, score, extra.args, debug = FALSE
 
     if (debug) {
 
-      cat("  > initial score for node", arc[1], "is", reference.score[arc[1]], ".\n")
+      cat("  > initial score for node", arc[2], "is", reference.score[arc[2]], ".\n")
       if (is.legal.arc(arc, score, data))
         cat("  > score delta for arc", arc[1], "->", arc[2], "is", better$delta, ".\n")
       else

R/choose.direction.R

@@ -44,8 +44,9 @@ custom.fit.backend = function(x, dist, discrete, ordinal) {
       # sanity check the distribution by comparing it to the network structure.
       if (is(dist[[node]]$coef, "matrix")) {
 
-        check.cgnode.vs.spec(dist[[node]], old = fitted[[node]]$parents,
-          node = node)
+        dist[[node]] =
+          check.cgnode.vs.spec(dist[[node]], old = fitted[[node]]$parents,
+            node = node, discrete = discrete[fitted[[node]]$parents])
         # set the correct class for method dispatch.
         class(fitted[[node]]) = "bn.fit.cgnode"
 
@@ -131,9 +132,6 @@ custom.fit.backend = function(x, dist, discrete, ordinal) {
         # identify discrete and continuous parents and configurations.
         parents = fitted[[node]]$parents
         configs = as.character(seq(from = 0, to = ncol(dist[[node]]$coef) - 1))
-        # store the new coefficients and standard deviations.
-        fitted[[node]]$coefficients = noattr(dist[[node]]$coef)
-        fitted[[node]]$sd = structure(noattr(dist[[node]]$sd), names = configs)
         # identify discrete and continuous parents.
         dparents = as.integer(which(discrete[parents]))
         gparents = as.integer(which(!discrete[parents]))
@@ -143,6 +141,9 @@ custom.fit.backend = function(x, dist, discrete, ordinal) {
         dlevels = sapply(parents[dparents],
           function(x) dimnames(dist[[x]])[[1]], simplify = FALSE)
         fitted[[node]]$dlevels = dlevels
+        # store the new coefficients and standard deviations.
+        fitted[[node]]$coefficients = noattr(dist[[node]]$coef)
+        fitted[[node]]$sd = structure(noattr(dist[[node]]$sd), names = configs)
         # reset columns names for the coefficients and names for sd.
         colnames(fitted[[node]]$coefficients) =
           names(fitted[[node]]$sd) = configs

R/custom.fit.R

@@ -1,7 +1,7 @@
 
 crossvalidation = function(data, bn, loss = NULL, k = 10,
-    m = ceiling(nrow(data)/10), algorithm.args, loss.args, fit, fit.args,
-    method, cluster = NULL, debug = FALSE) {
+    m = ceiling(nrow(data)/10), folds, algorithm.args, loss.args, fit,
+    fit.args, method, cluster = NULL, debug = FALSE) {
 
   n = nrow(data)
 
@@ -27,6 +27,14 @@ crossvalidation = function(data, bn, loss = NULL, k = 10,
     # all test sets have the same length, a dummy works just fine.
     kcv.length = rep(m, k)
 
+  }#THEN
+  else if (method == "custom-folds") {
+
+    # the folds are the custom folds specified by the user.
+    kcv = folds
+    # store the length of each test set.
+    kcv.length = sapply(kcv, length)
+
   }#THEN
 
   if (is.character(bn)) {

R/cv.R

@@ -47,7 +47,7 @@ fast.incremental.association = function(x, cluster = NULL, whitelist,
   nodes = names(x)
 
   # 1. [Compute Markov Blankets]
-  mb = smartLapply(cluster, as.list(nodes), fast.ia.markov.blanket,
+  mb = smartSapply(cluster, as.list(nodes), fast.ia.markov.blanket,
          data = x, nodes = nodes, alpha = alpha, B = B, whitelist = whitelist,
          blacklist = blacklist, test = test, debug = debug)
   names(mb) = nodes
@@ -56,7 +56,7 @@ fast.incremental.association = function(x, cluster = NULL, whitelist,
   mb = bn.recovery(mb, nodes = nodes, strict = strict, mb = TRUE, debug = debug)
 
   # 2. [Compute Graph Structure]
-  mb = smartLapply(cluster, as.list(nodes), neighbour, mb = mb, data = x,
+  mb = smartSapply(cluster, as.list(nodes), neighbour, mb = mb, data = x,
          alpha = alpha, B = B, whitelist = whitelist, blacklist = blacklist,
          test = test, debug = debug)
   names(mb) = nodes

R/fast-iamb.R

@@ -0,0 +1,22 @@
+fast.lm = function(data, node, parents, keep.fitted = FALSE) {
+
+  y = minimal.data.frame.column(data, node)
+
+  .Call("fast_lm",
+        data = data,
+        node = node, 
+        parents = parents,
+        keep = keep.fitted)
+
+}#FAST.LM
+
+fast.cglm = function(data, node, parents, configs, keep.fitted = FALSE) {
+
+  .Call("fast_cglm",
+        data = data,
+        node = node,
+        parents = parents,
+        configs = configs,
+        keep = keep.fitted)
+
+}#FAST.CGLM