jeraud / starclass

Stellar Classification

The TASOC Stellar Classification module

This module provides the stellar classification setup for the TESS Asteroseismic Science Operations Center (TASOC).

The code is available through our GitHub organisation (https://github.com/tasoc/starclass) and full documentation for this code can be found on https://tasoc.dk/code/.

Note

Even though the full code and documentation are freely available, we highly encourage users to not attempt to use the code to generate their own photometry or classifications from TESS. Instead we encourage you to use the fully processed data products from the full TASOC pipeline, which are available from TASOC and MAST. If you are interested in working on details in the processing, we welcome you to join the T'DA working group.

The overall strategy of the classification pipeline is to have different classifiers are run on the same data, and all the results from the individual classifiers are passed into an overall "meta-classifier" which will assign the final classifications based on the inputs from all classifiers.

Classification is done in two levels (1 and 2), where the first level separates stars into overall classes of stars that exhibit similar lightcurves. In level 2, these classes are further separated into the individual pulsation classes.

Installation instructions

• Start by making sure that you have Git Large File Storage (LFS) installed. You can verify that is installed by running the command:

  >>> git lfs version

• Go to the directory where you want the Python code to be installed and simply download it or clone it via git as:

  >>> git clone https://github.com/tasoc/starclass.git .
  

• All dependencies can be installed using the following command. It is recommended to do this in a dedicated virtualenv or similar:

  >>> pip install -r requirements.txt

How to run tests

You can test your installation by going to the root directory where you cloned the repository and run the command:

>>> pytest

How to run starclass

• The first step is to train the classifier. The classifier can be trained through the following command:

  >>> python run_training.py -tf 0.2

  The option -tf specifies the fraction of data to use for testing.

  The training set can be set through -t, by default this is currently set to keplerq9v3.

  The full list of possible arguments can be found in run_training.py.

  The actual classifiers together with their performance metrics are saved to ~/starclass/starclass/data/L1/{keplerq9v3}/

• Once the classifier has been trained, it can be applied to a new data set. The first step is to create a database that will list all the light curves that have to be classified, and to which the results and features will afterwards be saved.

  >>> python run_create_todo_list.py {/input/folder/with/all/the/light/curves/}

  In case the light curves are not stored in .fits or .fits.gz format, the regex pattern needs to be specified through --pattern.

  The full list of possible arguments can be found in run_create_todolist.py.

• Now that we have created an input database, we can actually run the classifier to classify all the light curves. In case you want to run starclass on a large scale multi-core computer, you need to make sure MPI is installed and also install the python library mpi4py.

  >>> python run_starclass.py {/input/folder/with/light/curves/}

  >>> mpiexec -n 4 python run_starclass_mpi.py {/input/folder/with/light/curves/}

  The full list of possible arguments together with additional information can be found in run_starclass.py and run_starclass_mpi.py.

  The results are all saved to the todo.sqlite file we created in the previous step.