prediction.R

#%%
library(LalRUtils)
load_or_install(c('tidyverse','magrittr', 'rio', 'data.table',
  'stargazer', 'mgcv', 'glmnet', 'caret', 'ranger', 'doMC', 'e1071',
  'parallel', 'tictoc', 'pushoverr'))
set.seed(42)

#registerDoMC(cores=4)
####################################################
rootdir = '/home/users/apoorval/tmp/FINAL/'
setwd(rootdir)
load('TrainData.RData')
colnames(train_cov) = paste0('pred', 1:ncol(train_cov))
traindat = cbind(train_dep, train_cov)
###############################################################
## setup
###############################################################
TrainingDataIndex <- createDataPartition(train_dep,
                                         p=0.75, list = FALSE)
# Create Training Data
trainingData <- traindat[TrainingDataIndex,]
testData     <- traindat[-TrainingDataIndex,]
# 10 fold CV
TrainingParameters <- trainControl(method = "repeatedcv",
                                   number = 10, repeats=10)
###############################################################
### RANDOM FOREST
###############################################################
Rand_Tuning <- trainControl(method = "repeatedcv",
                                   number = 10, repeats=10,
                                   # randomise tuning parameter search
                                   search = 'random')
tic()
rfModel <- train(train_dep ~ ., data = trainingData,
                 method = "ranger",
          # ranger parallelises on all cores by default
          # set to reasonable number
                 num.threads = 4,
                 trControl= Rand_Tuning,
                 metric = "RMSE",
                 tuneLength = 1e3,
                 #preProcess = c("scale","center"),
                 na.action = na.omit
)
toc()
RFPredictions <- predict(rfModel, testData)
RF_Diag = postResample(pred = RFPredictions,
                       obs = testData[, 1])
rfModel
RF_Diag
pushover_quiet('RF done')

###############################################################
#%% Produce Predictions
#load('TestCovs.Rdata')
# save.image('PredProb.Rdata')