DeepARTransit

DeepARTransit (Deep Auto-Regressive Transit) is a Python/Tensorflow package for de-trending transit light curves. It implements a stacked Long Short-Term Memory network nicknamed TLCD-LSTM (standing for Transit Light Curve Detrending LSTM), which:

• is trained to predict the next step's mean and standard deviation of a gaussian likelihood, in a similar fashion as in https://arxiv.org/abs/1704.04110
• rather than predicting in the future, the model is used for interpolating the input time-series on an inner chunk - typically on the in-transit time for transit light curves. No loss is computed on this range, where the input time-series values are substituted by the previous prediction (ancestral sampling).

Used on a transit lightcurve, the output of the TLCD-LSTM provides a prediction for the stellar+instrumental noise, from which one can then infer the obstructed flux due to the planetary transit by fitting a physical model to the detrended lightcurve.

Usage

1- Train model and provide in-transit flux predictions

• Pointing to an experiment folder containing a parameter file

$ python main_deepartrans.py -e experiment_folder_path 

• Pointing to a configuration file

$ python main_deepartrans.py -c configuration_file_path

2- Plot results and transit fitting

Example notebook for plotting and fitting results: post_processing.

3- Licensing and citing

This work is licensed under a Creative Commons Attribution 4.0 International License.

It can be cited using the following bibtex entry:

@misc{morvan_deepartransit_2019,
    author       = {Morvan, Mario},
    title        = {{DeepARTransit: A library for interpolating and detrending transit light curves with LSTMs.}},
    month        = Dec,
    year         = 2019,
    doi          = {190091225},
    version      = {1.1},
    publisher    = {Zenodo},
    url          = {https://zenodo.org/badge/latestdoi/190091225}
    }