Đang tải bảng tin…
Towards Data Science0 Hot0 bình luận12 phút đọc
My Model Was Cheating on Its Own Test
A used-car price prediction model built with scikit-learn's MLPRegressor originally reported a test R squared of 0.887, but the number was inflated by data leakage: outlier capping, StandardScaler, and OneHotEncoder were all fit on the full dataset before the train/test split, letting the model's preprocessing peek at test rows. After splitting first and fitting preprocessing only on training data, the honest test R squared dropped to 0.767, with mean squared error nearly quadrupling from about 6.9 million to 26.2 million. The piece walks through the buggy and corrected pipeline code, explains why each step looked innocent in isolation, and offers a checklist for avoiding this class of leakage in preprocessing pipelines.
Nguồn: https://towardsdatascience.com/my-model-was-cheating-on-its-own-test. 8 Sync News chỉ tóm tắt và dẫn link; bản quyền nội dung thuộc tác giả và nguồn gốc.
