search

IDEMSInternational / R-Instat

A statistics software package powered by R
http://r-instat.org/
GNU General Public License v3.0
38 stars 102 forks source link

Improving the Climatic > Model > Extremes dialogue #7968

Open rdstern opened 1 year ago

rdstern commented 1 year ago

I am making this issue from the pull request:

I have been working with @Vitalis95 and @Ivanluv on improving the extremes dialog. These are the improvements:

i) Under the Bayesian fit method, prior parameters, Initial values and number of iterations was added to improve the model. ii) We added the confidence intervals tab that outputs the parameter and return level, with corresponding return periods, confidence intervals. iii) We combined the fit method and display options sub dialogs into one sub dialog (Extremes Method) with separate tabs ie fit method, Display Options and Confidence Interval to make it easier for the user.

rdstern commented 1 year ago

RDS reaction to initial implementation by Tula: It will be good for you to explain, ready for testers, what you have done to test the new dialogue, before expecting others to check. This to include which data you used, and what models you fitted. a) I like the new Options sub-dialogue. It is a good improvement. Please tidy the bottom of the sub-dialogue, so the controls are centred and there isn't that large space. b) I tested with the dodoma data (i.e. a single site). I got the annual daily rainfall maximum which I used for the testing.

c) I tried the Bayesian fitting and it seems to work. I don't quite understand all the Bayesian options, so didn't change from the defaults that you supplied. Could you please give some descriptive details so I know what I can change. (I was fitting a gev.) I would like to alter aspects of the prior.

d) I then added the year so I am fitting a non-stationary model. It didn't work. I then found that the original didn't work either, and that's because it doesn't like missing values that are different for the dependent and independent variables. (We do need to be able to filter out missing values in the data.) This was easy in this case, as it was just the last 2 years. So I deleted them.

e) Then I tried again. I found the Bayesian doesn't work now. Possibly something needs to change in the priors. The ordinary MLE works fine. Please could you check this.

rdstern commented 1 year ago

Reply from Tula: The issues raised above have now been sorted. Attached is a description of how the added parameters in the dialogs work. I have used the Dodoma Dataset to test the dialogs.

Please have a look at it.

Testing the Extremes Dialogs..docx