Single transit events in Kepler
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
document
pbs
peerless
prediction
results
scripts
.gitignore
K01870.01_rho-prior.pdf
LICENSE
README.rst
environment.yml
ms_revised.pdf
peerless.pdf
setup.py

README.rst

peerless

The search for the transits of long-period exoplanets and binary star systems in the archival Kepler light curves.

Running the code

This project is (in theory) reproducible given enough compute time. Here are the steps:

Set up the environment & build extensions

To get started, you'll need to install miniconda3 and then install the peerless environment:

conda env create -f environment.yml
source activate peerless

where environment.yml is in the root of this repository.

There are two packages that you'll need to install following the specific installation instructions in their documentation: (a) the 1.0-dev branch of george, and (b) transit.

Once this environment is enabled, set the environment variable:

export PEERLESS_DATA_DIR="/path/to/scratch/"

to the directory where you want peerless to save all of its output. You'll need something like a TB of disk space to run the full pipeline.

Then, you'll need to build the peerless extensions:

python setup.py build_ext --inplace

Target selection & data download

Next up, run the target selection and download all the relevant datasets:

scripts/peerless-targets
scripts/peerless-datasets -p {ncpu}
scripts/peerless-download -p {ncpu}

where {ncpu} is the number of CPUs that you want to run in parallel using multiprocessing (they must be on the same node).

Transit search

To search these targets for transits, run:

scripts/peerless-search -p {ncpu} -q --no-plots -o {searchdir}

where {ncpu} is the same as above and {searchdir} is the root directory for the output.

Injection & recovery tests

Then to run a single pass of injection tests (one per target), run:

scripts/peerless-search -p {ncpu} -q --no-plots --inject -o {injdir}/{someinteger}

Since you'll want to run many rounds of this script, the output directory should be something like /path/to/injections/{someinteger} where {someinteger} is an integer identifying the run.

To collect the results of the search and injection tests, run:

scripts/peerless-collect {searchdir} {injdir} -o {resultsdir}

where {searchdir} and {injdir} are from above and {resultsdir} is the location where these should be saved. Some of the figure scripts will expect {resultsdir} to be results in this directory so, if you choose a different location, the figures might fail.

To predict the masses of the injected planets, run:

scripts/peerless-collect {searchdir} {injdir} -o {resultsdir}

MCMC sampling

To set up the MCMC fits for the candidates, run:

scripts/peerless-init {resultsdir}/candidates.csv -p -o {mcmcdir}

where {mcmcdir} is the directory where the MCMC results should be saved. Then, to run the MCMC analysis, run:

export NP={number_of_processes}
mpiexec -np $NP scripts/peerless-fit {mcmcdir}/{kicid}/init.pkl --nwalkers $((NP*2))

for each {kicid}. You'll probably want to rerun this script a few times to get more samples.

To collect the MCMC fit results, run:

scripts/peerless-collect-fits {resultsdir}/candidates.csv {mcmcdir} -o {resultsdir}

This script saves a table of posterior quantiles to {resultsdir}/fits.csv, figures to the directory document/figures for use in the manuscript, and HDF5 archives of thinned MCMC chains to {resultsdir}/chains.

False positive simulations & analysis

Run the predictions notebook. Dependencies are exosyspop, which further depends on isochrones and vespa.

Generate LaTeX tables and macros

Finally, to generate the LaTeX tables and macros for the paper, run:

scripts/peerless-write-tex {resultsdir}/candidates.csv {resultsdir}/fits.csv {resultsdir}/injections-with-mass.h5 {resultsdir}/fpp.csv

This will save several .tex files to the document directory.

License

Copyright 2015-2016 Daniel Foreman-Mackey

Licensed under the terms of the MIT License (see LICENSE).