-
Notifications
You must be signed in to change notification settings - Fork 231
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Nonuniform distributions for Morris sampling returns inf #515
Comments
Hi @jyangfsu Is the reference below the one you are referring to? I had a quick scan over it but could not see exactly where your suggestion is mentioned so I think you are referring to a different paper.
|
Hi, sorry for the confusion caused. See page 119 of this book:
http:https://www.andreasaltelli.eu/file/repository/A_Saltelli_Marco_Ratto_Terry_Andres_Francesca_Campolongo_Jessica_Cariboni_Debora_Gatelli_Michaela_Saisana_Stefano_Tarantola_Global_Sensitivity_Analysis_The_Primer_Wiley_Interscience_2008_.pdf
Also, 3.8 EXERCISES at page 128 gives an excellent example considering cutting the tails at quantiles 0.5 and 99.5% .
Regards,
Jing
…________________________________
From: Takuya Iwanaga ***@***.***>
Sent: Sunday, August 21, 2022 1:08 AM
To: SALib/SALib ***@***.***>
Cc: Jing Yang ***@***.***>; Mention ***@***.***>
Subject: Re: [SALib/SALib] Nonuniform distributions for Morris sampling returns inf (Issue #515)
Hi @jyangfsu<https://urldefense.com/v3/__https://github.com/jyangfsu__;!!PhOWcWs!wMvDbowbsCSc-fJz2NAwAq-YwRaSaNooTs6y_zQOeTVikVX8rscRk4SKArs7gy3TIYLVM_Tve6YwHDs_pF5WsntG$>
Is the reference below the one you are referring to? I had a quick scan over it but could not see exactly where your suggestion is mentioned so I think you are referring to a different paper.
Saltelli, A., P. Annoni, I. Azzini, F. Campolongo, M. Ratto, and S. Tarantola (2010).
"Variance based sensitivity analysis of model output. Design and estimator for the total sensitivity index."
Computer Physics Communications, 181(2):259-270,
doi:10.1016/j.cpc.2009.09.018.
—
Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/SALib/SALib/issues/515*issuecomment-1221469781__;Iw!!PhOWcWs!wMvDbowbsCSc-fJz2NAwAq-YwRaSaNooTs6y_zQOeTVikVX8rscRk4SKArs7gy3TIYLVM_Tve6YwHDs_pCm_9Mot$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/APBTNE6UMTB6XLOUJYUTCHLV2G2V3ANCNFSM53J2XBNA__;!!PhOWcWs!wMvDbowbsCSc-fJz2NAwAq-YwRaSaNooTs6y_zQOeTVikVX8rscRk4SKArs7gy3TIYLVM_Tve6YwHDs_pNhFQ4Fh$>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
Sorry, bit late to the party, but I just installed SALib (version 1.4.5, which is the newest available through
Or, to capture 95% of the normal distribution in the truncated version (instead of 90%, like in the code above):
|
Thanks for pointing this out. It's surprising to me, the system should have noticed the update on PyPi and make a PR to update to 1.4.6. I will make a PR to update. |
Thanks @tupui - I made a mental note to check on the conda feedstock after the 1.4.6 release but obviously it had left my mind when I woke in the morning. |
Woops, I just noticed that the modified code that I proposed last week doesn't give the results you'd expect: it returns values between the mean (b1) and the mean + 1x standard deviation (b2), instead of values ranging from (b1 - 2 x b2) to (b1 + 2 x b2), like you'd expect for a normal distribution. Not quite sure yet where exactly this goes wrong, have you looked at this in the meantime? |
Thanks @sitadrost , I'll try to make some time to have a look this weekend. |
Ah, of course, so silly of me:
|
Apologies for the long silence on this issue, short on time these days. I suspect that the suggested code will have implications for other sampling approaches, as the scaling is not used for just the Morris method. The suggested changes cause the test for the Oakley function to fail, as an example. This test uses LHS and expects all parameters to be normally distributed. I think in this case, would it be advisable to just direct users to specify That said, I'm not sure this is the "best" solution. For example, where to signpost not to use |
Hi, I am facing the same issue in the latest version. Has it been solved? |
Thanks to @jyangfsu for the issue submission.
When the parameters follows norm, lognorm and truncnorm distributions, nonuniform_scale_samples function would returns inf.
As suggested by Saltelli et al. (2010), this can be avoid by cutting the tails of , for example, the normal distributions, at quantiles 5 and 95%.
Code in current nonuniform_scale_samples function:
Suggest modifying to:
The text was updated successfully, but these errors were encountered: