Revisiting the Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC). Challenge Paper
- How to deal with missing values in time series?
- Can a Transformer-based network reach a reasonable classification score?
- How does the performance compare with some Seq2Seq or 1D convolution?
- The script Interpolate_and_Create_Dataset_GaussianProcess.py is the one that will generate the .h5 file needed to train the models. This script will take as input the different CSV needed from the Data/ folder. As it is written right now, it generates all the data (galactic + extragalactic). I will separate them into 3 different scripts if needed (one for the total data, one for galactic and another for extragalactic only).
- The script Interpolate_and_Create_Dataset_ZeroFilling.py does more or less the same except for the fact that the interpolation is not the Gaussian process but the "vanilla" one where we fill with zeroes.
- The script Obtain_Fourier_transform_from_Timeseries.py will read the .h5 generated by the other scripts and output only the data corresponding to the Fast Fourier Transform.
All the scripts generate different datasets within the .h5 file, where the separation into training, validation and test datasets is already considered.
- The script models.py contains the different classes from PyTorch defining the models, and utils.py contains miscellaneous stuff such as plotting functions, etc.
- The tests considered are contained in the Different_Tests/ folder, with a brief description of what has been done in the TXT file.