To reproduce the figures from the paper, please use define_experiment.ipynb
in the experiments folder.
All experiments have a definition for the data-model and problem types considered. The choice of the sweep parameters has to be customised.
Once the experiment has been defined, and a json file containing the data-model and sweep definition (usually called sweep_experiment.json
) been created, the data-models have to be created using the create_data_model.py
script.
Then, the experiment can be exectued using sweep.py
.
Running these scripts requires a working MPI installation. The command sequence is
mpiexec -n 5 python create_data_model.py sweep_experiment.json
mpiexec -n 5 python sweep.py sweep_experiment.json
Alternatively, in a cluster environment, it is possible to use the generated run.sh
file.
The sweep.py
script stores all results in a sqlite database. We provide scripts to easily extract the data in the Evaluate
folder.
The experiments on real data have been performed in the pca_experiments.ipynb
notebook.