Here we practice some regression techniques using a tabular dataset on house prices in Ames, Iowa.
Some key steps that were taken to produce the final predictions:
- Un-ordered categorical features were target encoded, rather than the more typically employed one hot encoding.
- Hyperparameter tuning was performed using a Bayesian hyperparameter optimization method.
- Three variants of gradient boosted trees are used: a vanilla gradient boosting regressor, light gradient boosting (light GBM), and extreme gradient boosting (XGB).
- A stacked meta-learner was also used.
- The final prediction is produced via the weighted average of predictions made using the 4 models listed above.
This notebook was executed on a Kaggle kernel, within the context of this ongoing Kaggle competition. If you'd like to fork this notebook with the associated data, please visit the version of this notebook that is hosted on Kaggle, available here.