This is the supplement for the submission "Nyström Kernel Stein Discrepancy" and contains the code used to reproduce the experimental results.
Our code depends on the code from (Jitkrittum et. al, 2017), available here, and that of (Huggins and Mackey, 2018), available here. For easy reference, we include both in the lib/ directory. We updated the latter to work with recent Python packages.
To reproduce the results of the experiments, please execute the following code (requires python>=3.11).
cd nystroem_ksd/
conda create -n ksd python=3.11
conda activate ksd
pip install ./lib/kernel-gof/
SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL=True pip install ./lib/random-feature-stein-discrepancies/
pip install -e .
sh scripts/run_all.sh
The notebooks/ folder contains the notebooks used for processing the results and for plotting.
Wittawat Jitkrittum, Wenkai Xu, Zoltan Szabo, Kenji Fukumizu, Arthur Gretton. A Linear-Time Kernel Goodness-of-Fit Test. NIPS, 2017.
Jonathan H. Huggins, Lester Mackey. Random Feature Stein Discrepancies. NIPS, 2018.