Code appendix for "UNIPoint: Universally Approximating Point Processes Intensities".
Simply run
pip3 install .
Download and unzip files into the data folder.
There are two sets of experiments available: (1) synthetic datasets; (2) real world datasets.
First train the models
python3 train_synth.py
Then evaluate for either log-likelihood and/or total variation
python3 eval_ll_synth.py
python3 eval_tv_synth.py
Resulting files of evaluation metrics per test sequence can be found in the eval folder.
Similar to synthetic, train the models first
python3 train_real.py
Then evaluate the log-likelihood
python3 eval_ll_real.py
The evaluated log-likelihood values are in the eval folder