search

usgs / pestpp

tools for scalable and non-intrusive parameter estimation, uncertainty analysis and sensitivity analysis
133 stars 71 forks source link

Parameter change summary file (pcs) #264

Open wkitlasten opened 1 year ago

wkitlasten commented 1 year ago

In the parameter change summary file (pcs) I'm getting 9 column headers: group,mean_change, std_change, num_at_near_lbound, percent_at_near_lbound, num_at_near_ubound ,percent_at_near_ubound, initial_cv, current_cv

but only 8 columns of data: drn_elev_layer1_199, 1867.12, 52.0759, 9, 0, 0, 37.8467, 1.06595

Comparision with the rec file suggests one of the ubnd results is missing?

version 5.2.5, compiled 29 June 2023

jtwhite79 commented 1 year ago

this is fixed on hotfix_mourestart and will be merged in to develop in the next couple of days...

wkitlasten commented 1 year ago

Thanks for fixing that. I can't for the life of me figure out why so many of my pars are at/near lower bounds, including ones I expect to go up. And NONE are ever at/near upper bounds. Is there a pyemu method by chance?

wkitlasten commented 1 year ago

Maybe I am looking at this the wrong way? I don't see much about pcs in the manual.

The attached zip contains a short notebook plus the files needed to explore my parubnd issue. Super psyched if someone can set me straight! ens_par_bnds.zip

jtwhite79 commented 1 year ago

Exception: Matrix.read_binary(): filename 'head_fixdrnelev_fixstage_balgp_0.0.par.jcb' not found

wkitlasten commented 1 year ago

Sorry, couldn't include both ensembles... thought I changed that to 'head_fixdrnelev_fixstage_balgp_0.1.par.jcb'. I think B found something.

wkitlasten commented 1 year ago

Hey, num_at_near_*bnd seems to be fixed, but all my % at *bnd fields are 0. Based on my calcs some of them should be above 0%. Any chance someone could check that calc at some point (sorry, I don't speak computer)? No rush, obviously.

jtwhite79 commented 1 year ago

@wkitlasten in an effort to limit output line widths, the percentages were being stored/written as ints. Any chance your expected percent at/near bound would be less than 1%? I'll convert them to floating points anyway, just curious if that is the issue...

wkitlasten commented 1 year ago

Better than int precision for percents might be useful, if not overkill. But it would also help if I multiplied by the correct number of realizations (facepalm). Thanks.