This is a repository containing some simple scripts which apply Tensorflow's sharding framework to statistical genetics models.
The sharding is done in runner.py, while the models themselves are specified in models.py.
To perform benchmarking, execute the master scripts as follows:
./[model]_benchmark.sh [n_GPU] [min_N] [max_N] [min_M] [max_M]
where the first parameter sets the number of GPUs made available to the pipeline, and the rest govern the range of exponents for the data sizes (samples and features, respectively). All results are stored in .hdf5 files in the data directory. Runtime estimates are also stored there in .txt files (with additional name component given by n_GPU).
Currently, the standard linear mixed model (lmm) and Border et al's assortative mating model (am) are available. More models may be implemented in the future.