Retrieving lower_ci, upper_ci for return periods

[mlicer]

Dear all,

i am sure there must be a simple way to retrieve lower_ci and upper_ci for return periods as two independent numpy arrays. I would like to do some custom plots and i don't seem to be able to obtain the confidence intervals...

I am currently doing something like:

...
return_periods = get_return_periods(
		ts=pd.Series(df[varname]),
		extremes=extremes,
		extremes_method=gevMethod,
		extremes_type="high",
		block_size=blockSize,
		return_period_size=blockSize,
		plotting_position="weibull",
	)

return_values = model.get_return_value(return_periods['return period'])

return eva_fit,extremes,return_periods, return_values

but all the returned return_values are nan...Any help would be appreciated!

Best regards,
M

[georgebv]

@mlicer This function returns a tuple with three arrays: return_value, ci_lower, and ci_upper as seen here:

pyextremes/src/pyextremes/eva.py

Lines 1036 to 1043 in f8b097f

	Returns
	-------
	return_value : array-like
	Return values.
	ci_lower : array-like
	Lower confidence interval bounds.
	ci_upper : array-like
	Upper confidence interval bounds.

You can do the following:

...
return_periods = get_return_periods(
		ts=pd.Series(df[varname]),
		extremes=extremes,
		extremes_method=gevMethod,
		extremes_type="high",
		block_size=blockSize,
		return_period_size=blockSize,
		plotting_position="weibull",
	)

reutrn_values, lower_ci, upper_ci = model.get_return_value(return_periods['return period'])

As for nans, I would need to see your full code but my gut feeling is that your blockSize is not 1 year - get_return_values assumes that return_period_size="1Y" and if it is not that then you can get nans when you ask for return period smaller than your block size in BM extremes.

[mlicer]

@georgebv Thank you very much for your reply and thanks for the library, i am sure it saved a lot of time to a lot of people! I have tried to modify the code along the lines of your suggestions but unfortunately i haven't been very successful. I've created a minimal example which reproduces the issue in this colab:

https://colab.research.google.com/drive/1q-HvazZvG6VB1Ii-so3_LIWvGi4Mr_2k?usp=sharing

If, by any chance, you can take a quick peek and execute the cells and let me know what i am doing wrong, i'd be most grateful!
Best regards!

[georgebv]

@mlicer I have created a gist for you here https://gist.github.com/georgebv/cdb3f2ad2f8bd8e0ebac37cef430468d

The reason you were getting nans was because you did not use the alpha parameter, which specifies confidence interval width. I should get it documented eventually.

[mlicer]

@georgebv thanks so much! This should clear things up. I truly appreciate it!
All the best, M