My attempt at the g2net hackathon in Thessaloniki.
| Dataset | Accuracy |
|---|---|
| Training | 87.53 % |
| Validation | 82.35 % |
| Test | 83.91 % |
optimazation.py: Contains the hyperparameter optimization using optuna.network.py: Contains the network used for the classification.helper.py: Contains helper functions for the network.submission.py: Contains the submission and evaluation function.
Run the optimization.py file to find the best hyperparameters for the network. Then run the network.py file to train the network. Finally, run the submission.py file to create a submission file and evaluate the network.
The data contain real noise segments of the Hanford detector during the O3 run, in which gravitational wave models were injected at different signal-to-noise ratios. The goal is to classify them into one of three categories:
- 0<=SNR<6
- 6 <=SNR <10
- SNR >= 10.
The injections correspond to binary black hole mergers with non-aligned spins, having masses between 7 and 50 times the solar mass, randomly distributed at different sky localizations and inclinations and in a distance ranges that results in signal to noise ratios up to about 50.
The full set of notebooks and tutorials can be found here.
