Merge pull request mi-erasmusmc#2 from mi-erasmusmc/development

First version of R package for EXPLORE including parallelisation

CMakeLists.txt

@@ -1,13 +1,20 @@
 cmake_minimum_required(VERSION 3.17)
 project(Explore)
 
-set(CMAKE_CXX_STANDARD 14)
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_FLAGS "-Wall -Wextra -g -O3")
 
 # set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
 # set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -v -stdlib=libc++")
 
+# MAC
 set(Boost_INCLUDE_DIR /Users/aniekmarkus/Documents/Documents/Code/C++/boost_1_74_0)
 set(Boost_LIBRARY_DIR /Users/aniekmarkus/Documents/Documents/Code/C++/boost_1_74_0/libs)
+
+# LINUX
+# set(Boost_INCLUDE_DIR /data/home/amarkus/Documents/boost_1_74_0)
+# set(Boost_LIBRARY_DIR /data/home/amarkus/Documents/boost_1_74_0/boost_1_74_0/libs)
+
 include_directories(${Boost_INCLUDE_DIR})
 link_directories(${Boost_LIBRARY_DIR})
 
@@ -17,6 +24,21 @@ include_directories(CMExplore/Debug)
 include_directories(Explore)
 include_directories(IOExplore)
 
+# MAC
+include_directories(include /usr/local/include/)
+link_directories(/usr/local/lib)
+# set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ltbb -fprofile-instr-generate -fcoverage-mapping")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ltbb --coverage")
+
+
+
+# LINUX
+# set(CMAKE_PREFIX_PATH "/opt/intel/oneapi/tbb/latest/lib/intel64/gcc4.8")
+# find_library(TBB_LIB tbb)
+# find_path(TBB_PATH
+ #       HINTS /opt/intel/oneapi/tbb/latest/include
+ #       NAMES tbb/parallel_for.h)
+
 set(SOURCE_FILES
         Clion/main.cpp
         src/C++/CMExplore/cmdline.h
@@ -81,5 +103,9 @@ set(SOURCE_FILES
 
 add_executable(Explore ${SOURCE_FILES})
 
+# LINUX
+target_link_libraries(Explore PRIVATE "${TBB_LIB}")
+target_include_directories(Explore PRIVATE "${TBB_PATH}")
+
 # add_library(ExploreLib SHARED ${SOURCE_FILES})
 # add_library(ExploreLib SHARED ${SOURCE_FILES})

DESCRIPTION

@@ -9,9 +9,9 @@ Description: This package includes the implementation of EXPLORE in C++ and conn
 License: GPL (>= 2)
 Depends:
     R (>= 3.6.1)
-Imports: Rcpp (>= 1.0.5)
+Imports: Rcpp (>= 1.0.5), RcppParallel
 Encoding: UTF-8
-LinkingTo: Rcpp, BH (>= 1.51.0)
+LinkingTo: Rcpp, BH (>= 1.51.0), RcppParallel
 RoxygenNote: 7.1.1
 Suggests: 
     testthat (>= 3.0.0),

NAMESPACE

@@ -8,4 +8,5 @@ export(runExplore)
 export(saveData)
 export(trainExplore)
 import(Rcpp)
+import(RcppParallel)
 useDynLib(Explore)

R/HelperFunctions.R

@@ -55,6 +55,8 @@ saveData <- function(output_path, train_data, file_name) {
   farff::writeARFF(train_data, paste0(output_path, file_name, ".arff"))
 
   # Create cutoff file based on column names
+  # TODO: add input to exclude class?
+  # FOR PLP: cutoff <- paste(colnames(train_data)[2:ncol(train_data)], collapse = ",")
   cutoff <- paste(colnames(train_data)[1:(ncol(train_data)-1)], collapse = ",")
   write.table(cutoff, 
               file=paste0(output_path, file_name,".cutoff"), 

R/MainFunctions.R

@@ -1,3 +1,4 @@
+# TODO: add descriptions for package manual
 
 #' trainExplore
 #'
@@ -35,6 +36,9 @@ trainExplore <- function(output_path, train_data = NULL, settings_path = NULL, f
     settings <- paste0(paste(readLines(settings_path), collapse="\n"),"\n")
   }
 
+  # TODO: check train_data and correct if necessary?
+  # drop nearly constant or extremely sparse values
+
   # Save train_data if entered
   if (!is.null(train_data)) {
     saveData(output_path, train_data, file_name)
@@ -74,10 +78,10 @@ trainExplore <- function(output_path, train_data = NULL, settings_path = NULL, f
 #'
 #' @return
 settingsExplore <- function(settings,
-                            output_path,
+                            output_path, # C++ cannot handle spaces in file path well, avoid those
                             file_name,
                             train_data = NULL,
-                            OutputFile = NULL,
+                            OutputFile = NULL, 
                             StartRulelength = NULL,
                             EndRulelength = NULL,
                             OperatorMethod = NULL,
@@ -91,7 +95,8 @@ settingsExplore <- function(settings,
                             PrintSettings = NULL,
                             PrintPerformance = NULL,
                             Subsumption = NULL,
-                            BranchBound = NULL) {
+                            BranchBound = NULL,
+                            Parallel = NULL) {
 
   # Insert location training data and cutoff file if train_data is entered
   if (!is.null(train_data)) {
@@ -106,16 +111,17 @@ settingsExplore <- function(settings,
   settings <- changeSetting(settings, parameter = "OutputFile", input = OutputFile, default_setting = paste0(output_path, file_name, ".result"))
   settings <- changeSetting(settings, parameter = "StartRulelength", input = StartRulelength, default_setting = 1)
   settings <- changeSetting(settings, parameter = "EndRulelength", input = EndRulelength, default_setting = 3)
-  settings <- changeSetting(settings, parameter = "OperatorMethod", input = OperatorMethod, default_setting = "ROCAREA")
+  settings <- changeSetting(settings, parameter = "OperatorMethod", input = OperatorMethod, default_setting = "EXHAUSTIVE")
   settings <- changeSetting(settings, parameter = "CutoffMethod", input = CutoffMethod, default_setting = "RVAC")
   settings <- changeSetting(settings, parameter = "FeatureInclude", input = FeatureInclude, default_setting = "")
   settings <- changeSetting(settings, parameter = "Maximize", input = Maximize, default_setting = "ACCURACY")
-  settings <- changeSetting(settings, parameter = "Accuracy", input = Accuracy, default_setting = "")
+  settings <- changeSetting(settings, parameter = "Accuracy", input = Accuracy, default_setting = "custom")
   settings <- changeSetting(settings, parameter = "Specificity", input = Specificity, default_setting = "")
   settings <- changeSetting(settings, parameter = "PrintSettings", input = PrintSettings, default_setting = "yes")
   settings <- changeSetting(settings, parameter = "PrintPerformance", input = PrintPerformance, default_setting = "yes")
-  settings <- changeSetting(settings, parameter = "Subsumption", input = Subsumption, default_setting = "no")
-  settings <- changeSetting(settings, parameter = "BranchBound", input = BranchBound, default_setting = "no")
+  settings <- changeSetting(settings, parameter = "Subsumption", input = Subsumption, default_setting = "yes")
+  settings <- changeSetting(settings, parameter = "BranchBound", input = BranchBound, default_setting = "yes")
+  settings <- changeSetting(settings, parameter = "Parallel", input = Parallel, default_setting = "no")
 
   # Save settings file
   settings_path <- paste0(output_path, file_name,".project")
@@ -139,6 +145,11 @@ settingsExplore <- function(settings,
 #' @examples
 predictExplore <- function(model, test_data) {
 
+  # Return NULL if cannot make predictions
+  if (is.null(model) || model == "") {
+    return(NULL)
+  }
+
   # Split string 
   all_terms <- stringr::str_split_fixed(model, "OR", n=Inf)
 

inst/settings/template.project

@@ -27,7 +27,7 @@ Accuracy=
 Specificity=
 [Output]
 OutputMethod=BEST
-PrintSettings=no
+PrintSettings=yes
 PrintPartitions=no
 PrintFeatureOperators=no
 PrintCutoffMethod=no
@@ -38,10 +38,11 @@ PrintCombinations=no
 PrintFeatureSets=no
 PrintCutoffSets=no
 PrintCutOffsetsBestLength=no
-PrintPerformance=no
+PrintPerformance=yes
 PrintSets=no
 SavePartitions=no
 [Run]
 OptimizeRules=yes
 Subsumption=yes
 BranchBound=yes
+Parallel=no

src/C++/Explore/candidate.cpp

@@ -7,6 +7,7 @@
 
 #include "candidate.h"
 #include "../common.h"
+#include <tbb/concurrent_vector.h>
 
 /**********************************************************************
 Function: Size()