gistfile1.txt

set.seed(998)
cl <- makePSOCKcluster(4)
registerDoParallel(cl)
## Create Model Data
arcmodel<-arc%>%dplyr::select(slaughtermonth, 
                       slaughteryear,
                       weightclasshalv,
                       length,
                       rochefant, 
                       `nir-ala`,
                       `nir-dha`,
                       `nir-dpa`,
                       `nir-epa`,
                       `nir-eta`,
                       generation1,
                       c.factor,
                       `nir-fasummon`,
                       `nir-fasumo3`,
                       `nir-fasumo6`,
                       `nir-fasumsatfa`,
                       `nir-linoleicacid`,
                       nirfat,
                       region, 
                       description,
                       customer)
glimpse(arcmodel)
arcmodel<-as.data.frame(arcmodel)
DataExplorer::plot_missing(arcmodel)
## Resolve Skewness of weight data
trans <- preProcess(arcmodel, method = c("BoxCox"))
arcmodel_trans <- predict(trans, arcmodel)
## Deal with outliers 
arcmodel_trans_trim<-as.data.frame(trim_df(arcmodel_trans, type="iqr"));
class(arcmodel_trans_trim)
## One Hot Encoding
dummy <- dummyVars(" ~ .", data=arcmodel_trans_trim)
arcmodel_trans_trim_dummy <- data.frame(predict(dummy, 
                                                newdata = arcmodel_trans_trim)) 
## Impute missing data 
arc.mis.knn.model = preProcess(arcmodel_trans_trim, 
                               method="knnImpute")
DataExplorer::plot_missing(arcmodel_trans_trim)
arcmodel_trans_trim_knn = predict(arc.mis.knn.model, 
                                  arcmodel_trans_trim) 
head(arcmodel_trans_trim_knn)
DataExplorer::plot_missing(arcmodel_trans_trim_knn)
dim(arcmodel_trans_trim_knn)
## Create Training and Testing datasets
inTraining <- createDataPartition(arcmodel_trans_trim_knn$rochefant, 
                                  p = .75, 
                                  list = FALSE)
training <- arcmodel_trans_trim_knn[ inTraining,]
testing  <- arcmodel_trans_trim_knn[-inTraining,]
dim(training)
dim(testing)

# Control Resampling Methods 
tr <- trainControl(method = "repeatedcv",
                   number = 40,
                   repeats = 20)

# Simple linear regression to get a feeling 
model<-lm(rochefant~slaughteryear + 
            slaughtermonth + 
            slaughteryear:slaughtermonth + 
            customer + description + 
            length + nirfat + region + generation1 + 
            weightclasshalv +
            weightclasshalv:length, 
          data=training) # what if the imputation messed up, regression to the mean? 
par(mfrow = c(2, 2))
plot(model)