Dependencies

• WebPPL
• python 3.x
• dendropy (I used version 4.4.0)
• nexus2phyjson
• pandas, seaborn (only for plotting)

The scripts will probably not work on Windows due to '/' vs. '' difference in path conventions.

Preparing data

Before running the python scripts, make sure you update the path to your nexus2phyjson installation in both scripts.

To generate synthetic data, run

python python/generate_trees.py

To convert the empirical data to phyjson, put the *tre files in data/empirical and run

python python/preprocess_real_trees.py

To run the benchmarks presented in the talk, run

python python/benchmark_on_tree.py

Running the main program

To run MCMC (Metropolis-Hastings), type

webppl bdInferMCMC.wppl --require . --require fs -- --tree path/to/phyjson/file [--burnin burnin --lag lag --samples nsamples] Where:

• burnin is the number of initial MCMC steps discarded as burn-in period (default: 5000)
• lag is the number of MCMC steps discarded between any two consecutive samples collected (default: 100)
• nsamples is the total number of samples collected (default: 300)

To run SMC, type

webppl bdInferSMCUnsorted.wppl --require . --require fs -- --tree path/to/phyjson/file [--rejuvSteps steps --samples nsamples]

Where:

• rejuvSteps is the number of rejuvenation steps at each iteratiion (default: 0, in all experiments rejuvSteps was at most 1)
• nsamples is the number of particles (default: 300)