Mission Overview

Convolutional Neural Networks for Flare Identification in TESS 2-minute Data ("stella")


Primary Investigator: Adina Feinstein

HLSP Authors: Adina Feinstein, Benjamin Montet, Megan Ansdell

Released: 2020-10-21

Updated: 2020-10-21

Primary Reference(s): Feinstein et al. 2020aFeinstein et al. 2020b

DOI: 10.17909/t9-7f44-6r51

Citations: See ADS Statistics

Read Me

Example of light curves colored by the average "probability" as determined by 10 ensembled stella CNN models. Flares are easily distinguishable from differing spot modulation behaviors.


Previous methods of flare detection with both Kepler and TESS data have relied on light curve detrending and using outlier detection heuristics for identifying flare events. stella is a novel way to detect flares in TESS short cadence data using convolutional neural networks (CNNs). Any TESS short cadence light curve can be run through the CNN models provided, without any detrending. The models created by the team return a probability light curve (see example figure), with values between 0-1 if a given light curve event is a flare or not. It takes < 1 minute to predict flares on a single TESS sector light curve using these models.

The CNN models were created with Google's machine learning API, Tensorflow. The team has created 100 trained ensembled models to use when predicting flares in other short cadence TESS light curves. Any single model can be used on its own, however the team recommends using at least 10 models and averaging the results. The details of each model can be found in Feinstein et al. 2020. The models can be opened and explored using either Tensorflow, h5py, or any other software that can open HDF5 files. An example of opening the models can be found using the Jupyter notebook in the Data Access section.

The stella software is installable through the Python package index via pip and can also be found on GitHub: https://github.com/afeinstein20/stella.  Users of the stella software or models should cite both of the Primary References linked at the top-left of this page.

Data Products

Data file naming convention:



  • <seed> = the random seed used to generate that model, which is a three-digit, zero-padded integer ranging from '000' to '099'.

Data file types:

_cnn.h5 Tensorflow CNN model.

Data Access

The Tensorflow CNN models can be downloaded from the table below, each link containing the HDF5 file for a given seed.  A tar bundle containing all of the HDF5 files can be downloaded here for those who want all the files (~100 MB): hlsp_stella_tess_ensemblemodel_all_tess_v0.1.0_bundle.tar.gz

A Jupyter notebook is available that demonstrates how to open and read in the models from the HDF5 files can be downloaded here: hlsp_stella_tess_ensemblemodel_any_tess_v0.1.0_demo.ipynb.

NOTE: this notebook is provided "as-is" by the HLSP team and MAST does not actively maintain this notebook.

000 001 002 003 004
005 006 007 008 009
010 011 012 013 014
015 016 017 018 019
020 021 022 023 024
025 026 027 028 029
030 031 032 033 034
035 036 037 038 039
040 041 042 043 044
045 046 047 048 049
050 051 052 053 054
055 056 057 058 059
060 061 062 063 064
065 066 067 068 069
070 071 072 073 074
075 076 077 078 079
080 081 082 083 084
085 086 087 088 089
090 091 092 093 094
095 096 097 098 099



Please remember to cite the appropriate paper(s) below and the DOI if you use these data in a published work. 

Note: These HLSP data products are licensed for use under CC BY 4.0.