These default tolerances are very small, checking the documentation for numpy reveals they are 1e-5 and 1e-8 for relative and absolute tolerance respectively. It would be useful to offer a way to configure these, either by command line argument or in the config.
Currently UPP requires flavour ratios to be close enough to eachother by using a numpy method with its default tolerances. https://github.com/umami-hep/umami-preprocessing/blob/main/upp/classes/components.py#L155
These default tolerances are very small, checking the documentation for numpy reveals they are 1e-5 and 1e-8 for relative and absolute tolerance respectively. It would be useful to offer a way to configure these, either by command line argument or in the config.