Fix predictive analystics module (#11)

* removed parallel processing

* removed multivariate option for test release

* added imputation of missing values

* fixed prediction plot display

The actual observations were not shown as first factor in prediction plot as they were a separate column from the predictions. This commit merges them in one column

* fixed bunch of stuff

* some bug fixes

- default values for forecast settings are now set based on data, previous values were set arbitrary and large and broke small dataset
- additionally pre-processing now only happens if at least one covariate is present - previously pre-processing threw an error when no covariates (only time variable) was provided
- additionally engineered lags and time-based variables are now in model dependency to recompute models once changed
- lastly lag values are now considered when forecast evaluation plan is created

DESCRIPTION

@@ -13,17 +13,15 @@ Imports:
   jaspBase,
   jaspGraphs,
   bsts,
-  parallel,
   bssm,
   precrec,
-  future,
-  future.apply,
   reshape2,
   Boom,
   lubridate,
   prophet,
   BART,
-  EBMAforecast
+  EBMAforecast,
+  imputeTS
 Suggests:
   testthat
 Remotes:

NAMESPACE

@@ -1,3 +1,2 @@
 import(jaspBase)
 export(predictiveAnalytics)
-export(multiVarControl)

R/modelFunctions.R

@@ -1,5 +1,5 @@
 #### bsts functions
-.bstsFitHelper  <- function(trainData,formula,method,niter=250,keepModel =F, ...){
+.bstsFitHelper  <- function(trainData,formula,method,niter=500,keepModel =F, ...){
   trainData$time <- as.numeric(trainData$time)
   ss <- list()
   y <- trainData[[all.vars(formula)[1]]]
@@ -44,7 +44,7 @@
 
 
   colnames(trainData)[colnames(trainData)=="time"] <- "ds"
-  m <- prophet::prophet(...,weekly.seasonality = F)
+  m <- prophet::prophet(...,weekly.seasonality = F,changepoint.range = 0.9)
   if(formula != as.formula("y ~ time")){
     for(reg in labels(terms(formula))[-1])
       m <- prophet::add_regressor(m,reg)
@@ -204,7 +204,8 @@
 
 
   reg_vars <- colnames(trainData)[colnames(trainData)!="y" & colnames(trainData) %in% labels(terms(formula))]
-  model <- xgboost::xgboost(data = as.matrix(trainData[,reg_vars]),nrounds = nrounds,nthread=1,
+  options(datatable.verbose=TRUE)
+  model <- xgboost::xgboost(data = as.matrix(trainData[,reg_vars]),nrounds = nrounds,nthread=1,verbose=0,
                             label = trainData$y,  metrics = list("rmse"), objective = "reg:squarederror",...)
 
 
@@ -231,7 +232,7 @@
 
   reg_vars <- colnames(trainData)[colnames(trainData)!="y" & colnames(trainData) %in% labels(terms(formula))]
 
-  model <- BART::wbart(x.train = trainData[,reg_vars],y.train = trainData$y,sparse = T,nkeeptreedraws = 100,ndpost = 100,nskip=100)
+  model <- BART::wbart(x.train = trainData[,reg_vars],y.train = trainData$y,sparse = T,nkeeptreedraws = 100,ndpost = 500,nskip=100)
 
   return(list(model=model))
 }