• Python 3.4+ or Python 3.3+ with enum package
• scipy/numpy
• Cython
• Statsmodels

• run fetch.sh in data/recombination_rates directory. This will fetch the appropriate recombination data from the HapMap project.
• run python3 setup.py build_ext --inplace in the predict directory. This will need to be run whenever any of the .pyx files are modified.

Notes:

• For the most part, executables are in the predict directory.
• Some of the script names don't make sense, because the files were named at a time when many functions were group together, then split out into other files.

The typical work flow is a three step process

1. Generate population - python3 generate_population.py --help
2. Gather samples - python3 run_classify_relationship.py --help
3. Identify - python3 evaluate_deanonymize.py --help

To generate a population use the generate_population.py script. For example if you cd into the predict directory and run: python3 generate_population.py ../data/tree_file ../data/recombination_rates/ --generation_size 1000 --num_generations 10 --output population.pickle A population with 10 generations each with 1000 members will be generated and saved to population.pickle with Python's pickle format.

To run experiments to collect data, run the command: python3 run_classify_relationship.py population.pickle work_dir 100 --num_labeled_nodes 150 --output_pickle distributions.pickle

This command will pick 150 nodes from the last generation and mark them as "labeled". All other nodes in the last 3 generations are considered "unlabeled". Then it will perform 100 experiments to sample from the simulated empirical distributions from the (labeled, unlabeled) pairs. After collecting 100 samples, it will fit a gamma distribution for each of these pairs to the empirical distribution.The total number of data points will be at most 100 samples multiplied by the number of (labeled, unlabeled) pairs.

Some pruning is done. Values are not sampled for pairs that are related more than --gen_back generations.

This command will delete the work_dir directory if it already exists, and create it if it doesn't (Unless the --recover option is used).

This command tends to take a long time to run. If this process is interrupted it can be resumed using --recover. If you provide the recover option, the --num_labeled_nodes option will be ignored, as the labeled nodes will be determined by work_dir. Recovering will try to do num_iteration new iterations, on top of what may already be in the work_dir. If num_iterations is 0, no experiments will be run, but rather the distributions will be calculated immediately.

The final step is identifying unlabeled individuals.

Running python3 evaluate_deanonymize.py population.pickle distributions.pickle -n 10 will try to identify 10 random unlabeled individuals in the population.