This project presents our ML solution for IJCAI 2018 Alimama pCTR competition.
run.ipynb: Jupyter Notebook of our solution.data_processing.py: Feature engineering part of our solution.- datasets: train and test datasets of the competition can be found here.
Following categories of features are extracted.
- User Features: features describing user properties (such as age, gender) and user behaviors (such as user's queries during a day/hour).
- Item Features: features describing item properties.
- Shop Features: features describing shop properties.
- Context Features: features describing context properties.
- Other Features: such as instance id, day, hour, etc.
Our algorithm is based on LightGBM, which is a fast, distrbuted and high performance gradient boosting framework based on decision tree algorithms. Specifically, we use the LGBMClassifier to efficiently train our model and predict on unknown instances.
Our solution achieved top 10% scores in the competition.