plt109/deficithawks

Adding my notes to the notebooks in the main repo

deficithawks

This is the code used to produce the results in Deficit hawks: robust new physics searches with unknown backgrounds.

• Figure 1 / Example experiment: cylinder.ipynb
• Figure 2 / Neyman construction: neyman_construction.ipynb
• Figure 3 / Region sets comparison: region_choice_plaw.ipynb
• Figure 4 / $Q_\alpha(s)$ : extra_scenarios.ipynb
• Figures 5 and 6 / Statistic comparison : statistic_choice.ipynb
• Figure 7 / Known background: known_bg.ipynb
• Figure 8 / Known and unknown background: mixed_bg.ipynb
• Figure 9 / Detection claims: discovery.ipynb
• Figure 11-14 / Additional Scenarios: extra_scenarios.ipynb

Installation

This code depends on the hypney inference package. Until this appears on pypi, you can install it via

pip install git+https://github.com/JelleAalbers/hypney

Afterwards, you can run the notebooks in the repository folder.

Caveats

If you want to play around with deficit hawks, this code may not be the ideal place to start. Although cylinder.ipynb can be helpful, the other notebooks mainly test different scenarios in bulk.

Instead, you may prefer to use your favourite inference software (whether that is RooFit, zfit, pyhf, etc.) to brew your own deficit hawk. All you need is the ability to:

• Do likelihood tests, and
• Estimate the distribution of a test statistic through toy Monte Carlo.

These are both necessary anyway to do likelihood ratio inference -- unless you're in one of the lucky situations where you can use an asymptotic result like Wilks' theorem without worry..