The benchmark has been realized on the FIFA dataset.
You can get the dataset with curl: curl http://www.philippe-fournier-viger.com/spmf/datasets/FIFA.txt --output FIFA.dat.
The training has been made with 20_450 sequences with an average length of 34 and an alphabet of 2990 elements.
The benchmark has been realized with a PC with 8 GB of ram, 8 cores and the Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz CPU.
With FIFA.dat in the data folder, you can run the benchmark from the root folder: python benchmark/benchmark.py.
Using multithreading, CPT made 2435 predictions per second, which is an average of 0.4 ms per prediction.
However, most use cases do not take advantage of multithreading. Without multithreading, CPT made 831 predictions per second, which is an average of 1.2 ms per prediction.