Practical Predictive Analytics

Test and train datasets

Of course, we will want to partition the data into test and training datasets. We have covered many ways to separate into test and train. This next method of partitioning will take a 80%/20% training/test split by:

• first extracting the index numbers from the training dataset. We will use the base::sample function to accomplish this
• building the training dataset from these indices
• constructing the test dataset from the rows which are not contained in the index.

Once we obtain the train indices, we will use the optbin function from the OneR package, which will optimally split a numeric variable based upon its ability to predict a Yes or No outcome for frisked. We have already seen this kind of optimal splitting with ...