This is less of a blog post and more of a data science life hack. I am all for making common routines easier, so when I stumbled upon this easy way to create train/test splits I thought I would share.
The package you will need is gradDescent
First, read in your dataset
library(data.table)
library(gradDescent)
df <- fread("your_dataset.csv")
gradDescent has a a great function called splitData()
Its arguments include:
- The dataset
- dataTrainRate
- seed
dataTrainRate controls what percentage of the data becomes training data and what percentage becomes test data.
So setting the dataTrainRate to 0.8 creates a 80/20 train/test split.
split_df= splitData(df, dataTrainRate = 0.8, seed = NULL)
Extract dataTrain and dataTest from the split_df and save them to varibles:
splitTrain = split_df$dataTrain
splitTest = split_df$dataTest
And that’s it! Just make sure you remove the column from the testset you want to predict on.
All in all, it’s a little less complicated than say:
set.seed(15)
index <- 1:nrow(df)
testindex <- sample(index, trunc(length(index)/3))
testset <- df[testindex,]
trainset <- df[-testindex,]