This repository contains the source code for Predict Student Performance from Game Play kaggle competition. Objective of the competition is to predict how well a student will score in quizes in an EdTech game based on their interactions during the gameplay.
This solution is ranked 143 out of 2103 (top 7%) in the private leaderboard and is placed in the Bronze medal category.
Ensure you have setup Kaggle API on your host, you may refer to this documentation on how to set it up.
Run the following code to download the competition data and divide the training data into individual level_groups.
python prepare_data.py
To train the models, run the following command
python train.py [--model] [--seed] [--folds] [--level_groups] [--threshold] [--n_trials]
Arguments:
--model Type of model to be trained. Valid options are 'lightgbm' and 'catboost'. Default is 'lightgbm'
--seed Seed value for reproducibility. Default is 42
--folds Number of folds for cross-validation. Default is 5
--level_groups Level group of the training data. Takes in the input as a list and accept either one of these values '0-4', '5-12', '13-22'
--threshold Threshold value used for binary classification. Default is 0.6
--n_trials Number of trials used in optuna hyperparameter optimization. Default is 1
Note that this competition uses the Kaggle Time-Series API during inference to ensure data is delivered in groupings that do not allow access to future data.
To perform prediction, run the following command
python infer.py [--models] [--lightgbm_wt] [--catboost_wt] [--folds] [--threshold]
Arguments:
--models Model for inference. Takes in the input as a list and valid options are 'lightgbm' and 'catboost'. Default is ['lightgbm', 'catboost']
--lightgbm_wt Weightage assigned to ensemble predictions for LightGBM. Default is 0.56
--catboost_wt Weightage assigned to ensemble predictions for CatBoost. Default is 0.30
--folds Number of folds used during training. Default is 5
--threshold Threshold value used for binary classification. Default is 0.53