Models diagnostics

Warning

This script constituted basically the benchmarks for the estimator models. It is still available, but its contents have been migrated to the benchmark notebooks and it is now progressively discontinued.

To get diagnostic plots in order to control the robustness and the performance of the models you can use the script model_diagnostic.py. It takes as arguments a configuration file:

usage: model_diagnostic.py [-h] --config_file CONFIG_FILE [--wave | --tail]

Make diagnostic plot

optional arguments:
  -h, --help            show this help message and exit
  --config_file         CONFIG_FILE
  --wave                if set, use wavelet cleaning
  --tail                if set, use tail cleaning, otherwise wavelets

For the energy estimator the diagnostic plots consist in:

• Distribution of the features

• Importance of the features

• Distribution of the ratio of the reconstructed energy over the true energy fitted with a gaussian for the subarrays

• Energy resolution and energy bias corresponding to the gaussian parametrisation for the subarrays

For a g/h classifier the following diagnostic are provided:

• Distribution of the features

• Importance of the features

• ROC curve (and its variation with energy)

• Output model distribution (and its variation with energy)