axtreme.qoi.marginal_cdf_extrapolation¶
MarginalCDFExtrapolation calculates QoIs by estimating the behaviour of a single timestep, and extrapolating.
Functions
|
Maximum possible single timestep error while still producing required accuracy in period estimate. |
|
Convert a long term quantile to an equivalent single timestep quantile. |
Classes
|
Estimate an ERD QoI using the marginal CDF for a single timestep. |
- axtreme.qoi.marginal_cdf_extrapolation.acceptable_timestep_error(q: float, period_len: int, atol: float = 0.01) float¶
Maximum possible single timestep error while still producing required accuracy in period estimate.
We often make estimates for a single timestep, and scale them to a period (many timesteps). e.g. q = q_timestep^period_len. Errors in the single timestep estimates compound. This function returns the largest error possible to a single timestep to still be within the required accuracy in the period estimate.
- Parameters:
q – long term quantile being estimated.
period_len – the number of timesteps that make up the period of interest.
atol – the maximum absolute error acceptable in q
- Returns:
The maximum absolute error acceptable in the q_timestep estimate.
- axtreme.qoi.marginal_cdf_extrapolation.q_to_qtimestep(q: float, period_len: int) float¶
Convert a long term quantile to an equivalent single timestep quantile.
- Parameters:
q – long term quantile being estimated.
period_len – the number of timesteps that make up the period of interest.
- Returns:
The equivalent quantile for a single timestep with error +-
Note
This funciton exists because there were numerical concerns for this process. Having a function allows us to document them in tests. It is appropriately accuract for very large periods. Periods of 1e13 creates an error of less that 1e-3 in q (the full quantile estimate). See test/qoi/test_margianl_cdf_extpolation/test_q_to_qtimestep_numerical_precision_period_increase for details.