GroupShuffleSplit
noun
say it: “GROOP SHUF-ul SPLIT” /ˈɡrup ˈʃʌfəl ˈsplɪt/
A scikit-learn tool that splits data into training and test parts while keeping every row of the same group, such as one customer, in the same part.
What is GroupShuffleSplit?
GroupShuffleSplit is a scikit-learn splitter that keeps all the rows of one group, such as one customer or one household, on the same side of the train-test split. When one customer is on both sides, the model is tested partly on people it already learned, and its score looks better than it will be.
from sklearn.model_selection import GroupShuffleSplit
visits = [[1], [2], [3], [4], [5], [6]]
customer = ["ana", "ana", "ben", "ben", "cy", "cy"]
splitter = GroupShuffleSplit(n_splits=1, test_size=1, random_state=0)
train, test = next(splitter.split(visits, groups=customer))
print(train, test) # [0 1 2 3] [4 5]: all of cy's visits in the test part
In a sentence
- “Switching to GroupShuffleSplit lowered the test score, because no customer was on both sides anymore.”
- “GroupShuffleSplit kept all of Ana's visits in the training rows.”
Level: Freelancer · More Python terms at this level
Updated