tabicl-survival 0.1.0

TabICL-Survival: A Tabular Foundation Model for Survival In-Context Learning

TabICL-Survival: A Tabular Foundation Model for Survival In-Context Learning

This repo is an adaptation of "TabICL: A Tabular Foundation Model for In-Context Learning on Large Data" for survival analysis. It is trained using Cox negative log likelihood.

Installation

Option 1: Installing tabicl-survival from PyPI

pip install tabicl-survival

Option 2: Installing tabicl-survival from the local clone

cd tabicl
pip install -e .

Option 3: Installing tabicl-survival directly from the git remote

pip install git+https://github.com/taltstidl/tabicl-survival.git

Basic Usage

from tabicl import TabICLSurver

surv = TabICLSurver()
surv.fit(X_train, y_train)  # this is cheap
surv.predict(X_test)  # in-context learning happens here

The code above will automatically download the pre-trained checkpoint (~325MB) from Hugging Face Hub on first use and choose a GPU if available. It supports datasets with up to 1,024 samples and 100 features at the moment.

Contributors

For the original TabICL implementation

• Jingang Qu
• David Holzmüller
• Marine Le Morvan

For the adapted TabICL-Surival implementation

• Thomas Altstidl