Changing pst.parameter_groups data?

[wkitlasten]

I am trying to change some of my parameter group names in both parameter_data and parameter_groups:

pst.parameter_data.loc[pst.parameter_data['pargp']==pg, 'pargp'] = npg
pst.parameter_groups.loc[pst.parameter_groups['pargpnme']==pg, 'pargpnme'] = npg

While debugging pst.parameter_data and pst.parameter_groups give the expected new names:

          parnme partrans parchglim  parval1   parlbnd  parubnd pargp  ...       pname  pstyle ptype  usecol    x    y zone
parnme                                                                 ...                                              
p0            p0      log    factor      1.0  0.666667      1.5   age  ...          tr       m    gr   input  NaN  NaN  NaN
p1            p1      log    factor      1.0  0.666667      1.5   age  ...          tr       m    gr   input  NaN  NaN  NaN
...

                  pargpnme    inctyp  derinc  derinclb  ...       splitthresh      splitreldiff splitaction extra
pargpnme                                                ...
porosity  porosity  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
kvmul        kvmul  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
kh              kh  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
age                age  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
cond              cond  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
flow              flow  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN

But when I do pst.write('new_file.pst', version=1) I get the old parameter groups:

* parameter groups
pg28                   relative              1.0000000000E-02      0.0 switch                2.0000000000E+00      parabolic             1.0000000000E-05      5.0000000000E-01      smaller              
pg31                   relative              1.0000000000E-02      0.0 switch                2.0000000000E+00      parabolic             1.0000000000E-05      5.0000000000E-01      smaller              
...

* parameter data
p0                    log                   factor                1.0000000000E+00      6.6666666667E-01      1.5000000000E+00      pg0                   1.0000000000E+00      0.0000000000E+00      1          
p1                    log                   factor                1.0000000000E+00      6.6666666667E-01      1.5000000000E+00      pg0                   1.0000000000E+00      0.0000000000E+00      1          
p2                    log                   factor                1.0000000000E+00      6.6666666667E-01      1.5000000000E+00      pg0                   1.0000000000E+00      0.0000000000E+00      1          
...

How can I get these new parameter group names to "stick" when I write the file?

[jtwhite79]

I think if you update the pargp names in parameter_data then call this method

pyemu/pyemu/pst/pst_handler.py

Line 1341 in add6dd4

def rectify_pgroups(self):

Then mod the par group info (not creating new records) it should stick..

[wkitlasten]

I should have mentioned, I am combining a bunch of old groups into the new groups. So after setting pargpnme to the new group using pst.parameter_groups.loc[pst.parameter_groups['pargpnme']==pg, 'pargpnme'] = npg within a loop for all old groups (pg) I do this:

pst.parameter_groups.drop_duplicates(inplace=True)
pst.parameter_groups.set_index('pargpnme', inplace=True, drop=False)

To get the df I want:

                  pargpnme    inctyp  derinc  derinclb  ...       splitthresh      splitreldiff splitaction extra
pargpnme                                                ...
porosity  porosity  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
kvmul        kvmul  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
kh              kh  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
age                    age  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
cond                  cond  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN
flow                  flow  relative    0.01       0.0  ...  1.0000000000e-05  5.0000000000e-01     smaller   NaN

With your suggestion I am doing this:

for pg in pst.parameter_groups.index:
    # some logic to find new group name = npg
    pst.parameter_data.loc[pst.parameter_data['pargp']==pg, 'pargp'] = npg
pst.rectify_pgroups()

This gives the same pst.parameter_groups as above. I'm not sure what you mean by "Then mod the par group info" so I am likely missing something.

Unfortunately pst.rectify_pgroups() does not solve my problem at this point and pst.write() still gives the old pargp/pargpnme names. It does miraculously update the parameter_groups df... I have no idea what manner of black magic it uses to align the old columns (inctyp, derinc, etc) with the new names!

Saving as version 2 gives the proper pargp in the csv files. So now I am just saving as v2, reloading, and saving as v1. Hacky, but it gives me what I need:

pst.write(os.path.join(model_ws, 'ensi.pst'), version=2)
pst = pyemu.Pst(os.path.join(model_ws, 'ensi.pst'))
pst.write(os.path.join(model_ws, 'ensi.pst'), version=1)

[briochh]

@wkitlasten, you shouldn't (need to) modify the pst.parameter_groups DataFrame directly. If you modify the pargp column of pst.parameter_data, pst.parameter_groups should get updated during the pst.write() step (effectively by running that pst.rectify_groups() line internally). So the steps are something like this:

• Modify groups for selected pars in paramter_data:

pst.paramter_data.loc[parindex_to_change, 'pargp'] = "newgroupname"

• [optional] Propagate changes to parameter_groups -- only necessary if you want to take a look at the changes before writing

pst.rectify_groups()

• Write control file

pst.write("filename.pst")

[wkitlasten]

Yep, not sure what I was missing... seems to work just fine now.