Skip to content

Latest commit

 

History

History
156 lines (107 loc) · 5.1 KB

README.rst

File metadata and controls

156 lines (107 loc) · 5.1 KB

Source for msprime paper

This repo contains all source code required to create the msprime paper.

Note that this is research software that was built for a particular purpose. It is not high quality software and probably contains bugs. Use at your own risk!

The src/algorithms.py file contains simple implementations of all the algorithms in the paper.

Prerequisites

First, clone the repo:

$ git clone https://github.com/jeromekelleher/msprime-paper.git

To run the code to compare simulations on a clean Ubuntu 15.10 system, we need the following:

$ sudo apt-get install python-dev python-pip libgsl0-dev libhdf5-serial-dev \
    pkg-config python-numpy python-pandas python-matplotlib libboost-dev \
    default-jre asymptote r-cran-ape python-biopython
$ sudo pip install -r requirements.txt

Paper and illustrations

To create the paper and illustrations, we first need to get the PLoS style files:

$ wget http://journals.plos.org/plosone/s/file?id=SyJ3/plos-latex-template.zip
$ unzip file\?id\=SyJ3%2Fplos-latex-template.zip

The location of this file will probably change, but these commands worked at the time of writing.

To compile the LaTeX document, run:

$ make

All of the prerequisites should already be installed, and the data required to create the plots is already present in the data directory.

Simulations

If you wish to run all the simulations for yourself, follow the three stages outlined in this section. However, if you just wish to obtain the plots from data used in the paper, run:

$ python src/plots.py plot all

All the plots should now be in the figures directory.

Running Simulations

Creating the plots to compare simulations works in three stages. First, we run the simulations and output the replicate data; second, we collect the replicate data and put it into a central file; and finally, we create the plots.

All work running simulations and plotting the results is done in the file src/plots.py.

First, we must download and build the other simulation packages that we'll be using:

$ make -C simulators

If everything goes according to plan, we should have working versions of the simulators we need once this completes.

Before we run any simulations, we must first set the locale to avoid problems with cosi2:

$ export LC_ALL=C

To run the simulations, use:

$ python src/plots.py simulate <dataset ID> -n <num_replicates> -p <num_processes>

The first argument to simulate is a dataset identifier. There are three datasets, alg_stats, tree_simulator_stats and haplotype_simulator_stats. There is also a special identifier all, which will run all available jobs sequentially. Therefore, we might run:

$ python src/plots.py simulate all -n 1 -p 8

This will run all available simulations with 1 replicate each (so, very inaccurate timings will be obtained) over 8 threads.

Obtaining accurate timing information from modern processors is quite challenging, as clock frequencies are dynamically rescaled constantly during workloads. This can lead to significant skews in timings if, for example, one timing was obtained when the processor core was running at peak frequency and another was obtained when it was at minimum frequency. A crude (but effective) way to work around this is ensure that the CPUs on the system are always busy when measurements are being taken. For example, if we have a machine with 20 threads then we first set off a stress job at low priority to ensure that all cores hit their equilibrium frequency:

$ nice -n 19 stress -c 20

This job will use up any spare CPU resources, and ensure that the clock frequencies of different cores don't vary unpredictably. If using an Intel processor it is also a good idea to turn off Turbo Boost while the measurements are being taken.

Processing results

Once a dataset has been simulated, we can then assemble it into a more convenient form for plotting. This is done by running:

$ python src/plots.py process <dataset ID>

In the same way as the simulate command, we provide a dataset ID as the argument, and we can pass the special all value to indicate that we want to process all data sets. After running the process command the raw replicate data in the data/results__NOBACKUP__ directory is compressed into the corresponding data files in the data directory. These are stored in JSON format for convenience.

The data used in the paper is included here in the data directory, and will be over written if you run process.

Plotting results

To plot the results, simply run:

$ python src/plots.py plot all

Similarly to the other commands, we provide a plot ID to the plot command to determine which figure we should draw.

Examples

To run the examples in which we compare the sizes of files in various formats and benchmark different Newick parsers, run:

$ make -C examples