Codes for paper 'Htgtm: Hybrid Temporal-Graph Tabular Model For Complex Multimodal Tabular Data Processing' (http://dx.doi.org/10.1109/MLSP55844.2023.10285918)
Context:
The paper is the winner of IEEE MLSP 2023 Data Challenge: Volunteer Retention (https://www.kaggle.com/competitions/MLSP2023-volunteers-01).
To run the codes:
Please place the following files in this folder: "retention_train_feat.csv", "retention_test_feat.csv" and "retention_gt_train.csv"
The files "gdforeign_ma.csv" and "gdlocal_ma.csv" contain the 7-day moving average of Guangdong's imported COVID-19 cases and locally transmitted COVID-19 cases, respectively.
"user_id_remapping.pkl" contains a mapping from user_id to data index. Additionally, "train_index.pkl" and "val_index.pkl" are used to split the original training dataset into a new training set and a holdout validation set.
Please use "htgtm.ipynb" to obtain an evaluation on the holdout validation set.