These programs implement the algorithm presented in our blog post, Synthesizing Realistic Timestamps. There are two many programs, analyze.py and generate.py. The only requirement is Python 3.5+.

analyze.py takes a sorted list of timestamps and fits an adaptive, non-homogeneous sequence of exponential distributions to that data. The output is two files, out_lambdas.dat and out_boundaries.dat that are meant to be used with generate.py. Here's an example of running analyze.py.

$ python analyze.py --start 2019-02-04 --end 2019-02-05 test_data.dat

--start and --end are the domain of the input data. test_data.dat is a sample of one day of events provided as an example.

generate.py takes the output of analyze.py and a time range, and synthesizes events with that model. Here's an example of running generate.py

$ python generate.py --start 2019-01-01 --end 2019-01-30 out_lambdas.dat out_boundaries.dat

Notice that the --start and --end times don't need to overlap with original data. You can generate as much data as you want from the model.