/astro-tl

transfer learning in astronomical spectroscopy

Primary LanguagePython

Transfer Leaning in Astronomical Spectroscopy

Use of transfer learning (in the context of deep learning) to classify QSOs in LAMOST (target domain) using SDSS (source domain) as the training data.

Envrinment Setup

The envrinment is based on Python 3.7.2, CUDA 10.1 and Parallel HDF5.

$ module load Python/3.7.2-fosscuda-2019a
$ module load HDF5/1.10.5-gompic-2019a
$ virtualenv venv    # create a virtual environment
$ source venv/bin/activate    # activate the environment
$ # install h5py with MPI mode enabled
$ pip install mpi4py
$ CC="mpicc" HDF5_MPI="ON" HDF5_DIR=/path/to/parallel-hdf5 pip install --no-binary=h5py h5py
$ pip install torch==1.5.0+cu101 torchvision==0.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html
$ pip install -r requirements.txt