TabSurv

This repository contains the official implementation of TabSurv, proposed in the accompanying paper.
The code is provided for reproducibility and research purposes.

Environment Setup

The repository includes an environment.yml file for creating a Conda environment with all required dependencies.

conda env create -f environment.yml
conda activate tabsurv

Repository Structure

The repository is organized as follows:

.
├── environment.yml
├── data/
│   └── *.npz
├── parts_ablation/
│   ├── *.json
│   └── parts_ablation.ipynb
├── pkg/
│   ├── model/
│   │   ├── *.py
│   ├── experiments/
│   │   ├── models/
│   │   ├── results/
│   │   ├── losses.csv
│   │   ├── paper_format.py
│   │   ├── simulation.py
│   │   ├── experiment.py
│   │   └── *.py
│   └── params_selection/
│       ├── params/
│       └── *.py

data/

Contains compressed datasets (.npz) used in the experiments reported in the paper.

parts_ablation/

Contains experiments for Figure 4 in the paper.

pkg/model/

Implementation of the proposed TabSurv method.

pkg/experiments/

Code for running experiments and reproducing results from the paper.

• experiment.py — main entry point for running experiments.
• models/ — source code and adapters for baseline models used for comparison.
• results/ — directory where experimental results are stored.
• simulation.py — code for generating the synthetic dataset used in experiments.
• paper_format.py — utilities for aggregating results and producing tables and figures as reported in the paper.

pkg/params_selection/

Scripts for hyperparameter optimization using Optuna.

• Each script corresponds to a specific model.
• params/ contains the selected hyperparameters used in all reported experiments.

Running Experiments

All experiments are executed from the root of the repository using Python module execution.

For example, to run the experiment corresponding to Table 3 on the SEER dataset:

python -m pkg.experiments.experiment -data SEER -device cuda:1

After the experiments are completed, the results can be aggregated and formatted using:

python -m pkg.experiments.paper_format

Hyperparameter Selection

Hyperparameters for each model are selected using Optuna.

For example, to perform hyperparameter optimization for the TabSurv_LS(MLP) model on the SEER dataset:

python -m pkg.params_selection.mlp -data SEER

All selected hyperparameters used in the experiments are stored in:

pkg/params_selection/params/