How to prepare features and labels for XGBoost XGBRanker?

Question

I'm going to use XGBRanker to make a recommender system. But in the official docs I haven't found an example of how to prepare the dataset.

So, in which format features and labels should be fitted in XGBRanker?

Answer 1

We need the below dataset for XGBRanker:

1. X as features
2. Y as target value(label)
3. group as kwargs(unique groups defined as per the domain) / qid per row

Make a note that group information should be an array consisting of the # of elements per group.

For example:

Groups in data: 4

Training, Test: (100,50)

Training Data(X,Y): (100,3)[feature 1, feature 2, feature 3]

Training Label(Y): (100,1)

Training group[A,B,C,D] i.e **grpInfo**: [25,25,10,35]

Note:The sum of grpInfo should be equal to # of records in TrainingSet i.e in this case this is 100

XGBRanker(X[[feature 1, feature 2, feature 3]],Y[[Label]], group=grpInfo)