EarlyStopping#

EarlyStopping is a Python library implementing computationally efficient model selection methods. For iterative estimation procedures applied to statistical learning problems, it is necessary to choose a suitable iteration index to avoid under- and overfitting. Classical model selection criteria can be prohibitively expensive in high dimensions. Recently, it has been shown for several regularisation methods that sequential early stopping can achieve statistical and computational efficiency by halting at a data-driven index depending on previous iterates only.

pip install EarlyStoppingPy
python3 -m pip install build virtualenv               # Install build tools
git clone https://github.com/ESFIEP/EarlyStopping.git # Clone git repository
python3 -m build                                      # Build package
python3 -m venv myenv                                 # Create virtual environment
source myenv/bin/activate                             # Activate virtual environment
python3 -m pip install numpy ipykernel                # Install python packages to the environment
python3 -m pip install -e .                           # Install the EarlyStopping package in editable mode
python3 -m ipykernel install --user --name=myenv      # Create Jupyter kernel from the environment
EarlyStopping Animation

References

Supported Methods

  • Truncated SVD

  • Landweber Algorithm

  • Conjugate Gradient Descent

  • L2 Boosting

  • Regression Tree (CART)

This research has been partially funded by the Deutsche Forschungsgemeinschaft (DFG) – Project-ID 318763901 - SFB1294, Project-ID 460867398 - Research Unit 5381 and the German Academic Scholarship Foundation.