Merge branch 'master' into fix201_focussearch

Author: jakob-r

.gitignore

@@ -10,4 +10,5 @@ bench/mlr/modelsel-files
 src/*.o
 src/*.so
 plots
-tests/testthat/Rplots.pdf
+tests/testthat/Rplots.pdf
+*.sublime-*

.travis.yml

@@ -1,17 +1,46 @@
 language: R
 sudo: false
-cache: packages
-warnings_are_errors: false
+r:
+  - release
+addons:
+  apt:
+    sources:
+      - ubuntu-toolchain-r-test
+    packages:
+      - gcc-4.9
+      - g++-4.9
+      - gfortran-4.9
+      - libgmp-dev
 env:
   global:
     - _R_CHECK_TIMINGS_=0
+    - _R_CHECK_FORCE_SUGGESTS_=0
     - secure: "e2QLomaUqNvpHnGNdBXS4VO2/UWokrKl9UgYrqWo+fhpXaJiJuONNPNL44BuWZR2Cy7noTHrevVbTKBkGpqJR42mkjzcjd6qyR1Ctiveir+84/HTtmexouIUn5OFRH5dgyt/gquld31RlHlSYoKho3nZ7D7SnVExov9tp2FGRhU="
 
+warnings_are_errors: false
+r_check_args: "--as-cran --run-donttest"
+
 r_github_packages:
-  - berndbischl/ParamHelpers
+  - berndbischl/parallelMap
   - jakobbossek/smoof
   - mlr-org/mlr
 
+r_packages:
+  - roxygen2
+
+
+before_install:
+  - mkdir $HOME/bin
+  - ln -s $(which gcc-4.9) $HOME/bin/gcc
+  - ln -s $(which g++-4.9) $HOME/bin/g++
+  - ln -s $(which gfortran-4.9) $HOME/bin/gfortran
+  - export PATH=$HOME/bin:$PATH
+  - echo $LD_LIBRARY_PATH
+  - echo $LIBRARY_PATH
+
+after_failure:
+  - ./travis-tool.sh dump_logs
+
 notifications:
   email:
     recipients:

NAMESPACE

@@ -41,9 +41,12 @@ import(BBmisc)
 import(ParamHelpers)
 import(checkmate)
 import(ggplot2)
+import(grDevices)
 import(lhs)
 import(mlr)
 import(parallelMap)
 import(smoof)
+import(stats)
+import(utils)
 useDynLib(mlrMBO,c_eps_indicator)
 useDynLib(mlrMBO,c_sms_indicator)

R/OptProblem.R

@@ -32,11 +32,10 @@ NULL
 # All variables in this Object should be documented here.
 # Think of it, when you implement new ones!
 # Unfortunately in R we cannot hinder you from putting other values in this object, but please: Don't!
-makeOptProblem = function(fun, par.set, design = NULL, learner, control, show.info = TRUE, more.args = list()) {
+makeOptProblem = function(fun, design = NULL, learner, control, show.info = TRUE, more.args = list()) {
   opt.problem = new.env()
 
   opt.problem$fun = fun
-  opt.problem$par.set = par.set
   opt.problem$design = design
   opt.problem$learner = learner
   opt.problem$control = control
@@ -62,7 +61,7 @@ getOptProblemControl = function(opt.problem) {
 }
 
 getOptProblemParSet = function(opt.problem) {
-  opt.problem$par.set
+  getParamSet(getOptProblemFun(opt.problem))
 }
 
 getOptProblemMoreArgs = function(opt.problem) {

R/checkLearner.R

@@ -16,7 +16,7 @@ NULL
 # check and create default learner
 checkLearner = function(learner, par.set, control) {
   if (missing(learner) || is.null(learner)) {
-    if (!hasDiscrete(par.set)) {
+    if (!hasDiscrete(par.set, include.logical = TRUE)) {
       if (control$noisy) {
         learner = makeLearner("regr.km", covtype = "matern5_2", predict.type = "se", nugget.estim = control$noisy)
       } else {

R/convertOptPathToDf.R

@@ -7,7 +7,7 @@
 #   MBO control object.
 # @return [\code{data.frame}]
 convertOptPathToDf = function(opt.path, control) {
-  df = as.data.frame(opt.path, discretes.as.factor = TRUE, include.rest = FALSE)
+  df = as.data.frame(opt.path, include.rest = FALSE)
   df = convertDataFrameCols(df, ints.as.num = TRUE, logicals.as.factor = TRUE)
   return(df)
 }

R/doc_mbo_OptPath.R

@@ -18,7 +18,7 @@
 #'   }
 #' }
 #' \item{multipoint.cb.lambda}{Random lambda-value used in q-CB point proposal. One lambda for each point in that case.}
-#' \item{.weight}{Weight vector sampled for multipoint ParEGO}
+#' \item{parego.weight}{Weight vector sampled for multipoint ParEGO}
 #' }
 #'
 #' Moreover, the user may pass additional \dQuote{user extras} by appending a named list

R/exampleRun.R

@@ -38,6 +38,8 @@ exampleRun = function(fun, design = NULL, learner = NULL, control,
   par.types = getParamTypes(par.set)
   n.params = sum(getParamLengths(par.set))
   noisy = isNoisy(fun)
+  control$noisy = noisy
+  control$minimize = shouldBeMinimized(fun)
   learner = checkLearner(learner, par.set, control)
   assertClass(control, "MBOControl")
   points.per.dim = asCount(points.per.dim, positive = TRUE)
@@ -51,8 +53,6 @@ exampleRun = function(fun, design = NULL, learner = NULL, control,
     global.opt = smoof::getGlobalOptimum(fun)$value
   else
     global.opt = NA_real_
-  control$noisy = noisy
-  control$minimize = shouldBeMinimized(fun)
 
   if (control$n.objectives != 1L)
     stopf("exampleRun can only be applied for single objective functions, but you have %i objectives! Use 'exampleRunMultiCrit'!",
@@ -80,9 +80,9 @@ exampleRun = function(fun, design = NULL, learner = NULL, control,
   if (is.null(fun.mean)) {
     evals = evaluate(fun, par.set, n.params, par.types, noisy, noisy.evals, points.per.dim, names.x, name.y, seq_along(control$multifid.lvls))
   } else {
-    evals = evaluate(fun.mean, par.set, n.params, par.types, noisy = FALSE, noisy.evals = 1, points.per.dim, names.x, name.y, seq_along(control$multifid.lvls))
+    evals = evaluate(fun.mean, par.set, n.params, par.types, noisy = FALSE, noisy.evals = 1, points.per.dim, names.x, name.y, multifid.lvls = seq_along(control$multifid.lvls))
   }
-
+  
   if (is.na(global.opt))
     global.opt.estim = ifelse(shouldBeMinimized(fun), min(evals[, name.y]), max(evals[, name.y]))
   else
@@ -93,7 +93,7 @@ exampleRun = function(fun, design = NULL, learner = NULL, control,
     messagef("Performing MBO on function.")
     if (is.null(design))
       messagef("Initial design: %i. Sequential iterations: %i.", control$init.design.points, control$iters)
-    messagef("Learner: %s. Settings:\n%s", learner$id, mlr:::getHyperParsString(learner))
+    messagef("Learner: %s. Settings:\n%s", learner$id, mlr:::getHyperParsString(learner, show.missing.values = FALSE))
   }
 
   # run optimizer now
@@ -104,7 +104,14 @@ exampleRun = function(fun, design = NULL, learner = NULL, control,
   if (!is.null(fun.mean)) {
     y.true = vnapply(convertRowsToList(getOptPathX(res$opt.path), name.list = TRUE, name.vector = TRUE), fun.mean)
   }
-
+  
+  if (control$multifid) {
+    n.params = n.params - 1
+    par.set = dropParams(par.set, ".multifid.lvl")
+    par.types = getParamTypes(par.set)
+    names.x = getParamIds(par.set, repeated = TRUE, with.nr = TRUE)
+  }
+  
   makeS3Obj("MBOExampleRun",
     fun.mean = fun.mean,
     par.set = par.set,
@@ -135,7 +142,7 @@ print.MBOExampleRun = function(x, ...) {
   catf("True points per dim.        : %s", collapse(x$points.per.dim))
   print(x$control)
   catf("Learner                     : %s", x$learner$id)
-  catf("Learner settings:\n%s", mlr:::getHyperParsString(x$learner))
+  catf("Learner settings:\n%s", mlr:::getHyperParsString(x$learner, show.missing.values = FALSE))
   mr = x$mbo.res
   op = mr$opt.path
   catf("Recommended parameters:")
@@ -158,29 +165,13 @@ getEvals = function(fun, par.set, noisy, noisy.evals, points.per.dim, names.x, n
   return(evals)
 }
 
-getEvalsForMultifid = function(fun, par.set, noisy, noisy.evals, points.per.dim, names.x, name.y, multifid.lvls) {
-  eval.list = lapply(multifid.lvls, function(l){
-    force(l)
-    force(fun)
-    this.fun = function(x) {
-      x$.multifid.lvl = l
-      fun(x)
-    }
-    evals = getEvals(this.fun, par.set, noisy, noisy.evals, points.per.dim, names.x, name.y)
-    cbind.data.frame(.multifid.lvl = l, evals)
-  })
-  do.call("rbind", eval.list)
-}
-
 evaluate = function(fun, par.set, n.params, par.types, noisy, noisy.evals, points.per.dim, names.x, name.y, multifid.lvls = numeric(0)) {
   if (!noisy && n.params == 1L && par.types == "discrete")
     stopf("ExampleRun does not make sense with a single deterministic discrete parameter.")
+  if (length(multifid.lvls) && n.params %in% c(2L,3L) && all(par.types %in% c("numeric", "numericvector", "discrete", "integer")))
+    return(getEvals(fun, par.set, noisy, noisy.evals, points.per.dim * length(multifid.lvls), names.x, name.y))
   if (n.params %in% c(1L, 2L) && all(par.types %in% c("numeric", "numericvector", "discrete"))) {
-    if (length(multifid.lvls)) {
-      return(getEvalsForMultifid(fun, par.set, noisy, noisy.evals, points.per.dim, names.x, name.y, multifid.lvls))
-    } else {
-      return(getEvals(fun, par.set, noisy, noisy.evals, points.per.dim, names.x, name.y))
-    }
+    return(getEvals(fun, par.set, noisy, noisy.evals, points.per.dim, names.x, name.y))
   }
 }
 

R/exampleRunMultiCrit.R

@@ -85,7 +85,7 @@ exampleRunMultiCrit= function(fun, design = NULL, learner, control, points.per.d
 
   showInfo(show.info, "Performing MBO on function.")
   showInfo(show.info, "Initial design: %i. Sequential iterations: %i.", control$init.design.points, control$iters)
-  showInfo(show.info, "Learner: %s. Settings:\n%s", learner$id, mlr:::getHyperParsString(learner))
+  showInfo(show.info, "Learner: %s. Settings:\n%s", learner$id, mlr:::getHyperParsString(learner, show.missing.values = FALSE))
 
   # run optimizer now
   res = mbo(fun, design, learner = learner, control = control, show.info = show.info)
@@ -145,7 +145,7 @@ print.MBOExampleRunMultiCrit = function(x, ...) {
   catf("Parameter types               : %s", collapse(x$par.types))
   print(x$control)
   catf("Learner                       : %s", x$learner$id)
-  catf("Learner settings:\n%s", mlr:::getHyperParsString(x$learner))
+  catf("Learner settings:\n%s", mlr:::getHyperParsString(x$learner, show.missing.values = FALSE))
   catf("Hypervolume                   : %.4e", x$mbo.hypervolume)
   catf("NSGA2 Hypervolume (6000 FEs)  : %.4e", x$nsga2.hypervolume)
 }

R/getExtras.R

@@ -13,7 +13,7 @@
 # errors.model            character(1)
 # filter.replace          logical(1)
 # multipoint.cb.lambda    numeric(1)
-# .weight<j>              numeric(1)
+# parego.weight.<j>         numeric(1)
 #
 # Please document the content in doc_mbo_OptPath.R
 
@@ -54,7 +54,7 @@ getExtras = function(n, prop, train.time, control) {
       weight.mat = prop$weight.mat
       if (is.null(weight.mat))
         weight.mat = matrix(NA_real_, nrow = n, ncol = control$n.objectives)
-      w = setNames(as.list(weight.mat[i, ]), paste0(".weight", 1:ncol(weight.mat)))
+      w = setNames(as.list(weight.mat[i, ]), paste0("parego.weight.", 1:ncol(weight.mat)))
       ex = c(ex, w)
     }
     # if we filtered proposed points, store flag