Dataset generation and ML scripts for localization of narrow-band signals, as described in Brzycki et al. 2020 (PASP).
create_dataset.py
: Dataset generation script, produces entirely synthetic data frames with ideal chi-squared background noise and constant intensity narrow-band signals. Creates both one and two signal datasets. Uses Setigen to create synthetic frames.
train_cnn.py
: Contains all ML-related code, including model architectures, custom data generators, and training/testing routines. Accepts command-line arguments to facilitate multiple experiments. Uses Keras with a Tensorflow backend.
run_training.sh
: Bash script executing train_cnn.py
for the full set of experiments, as they appear in the paper.
turboseti_analysis.py
: Uses TurboSETI on one signal test data to generate localizations, for comparison with ML model predictions.
Frame_generation.ipynb
: Jupyter notebook illustrating some basic data frame generation using setigen
.
RMSE_figures.ipynb
: Jupyter notebook producing the RMSE plots found in the paper, using the output from test predictions via train_cnn.py
.