saving params from minimizer result??

1,371 views
Skip to first unread message

김국태

unread,
Jun 28, 2019, 6:05:39 AM6/28/19
to lmfit-py
Dear lmfit users & developers.

I have a simple code like below.

# code start
import numpy as np
import matplotlib.pyplot as plt
import lmfit as lmf
from math import pi

mu = 12.57E-7;
R = 0.0125; #Radius of metal sphere(m)
m = 0.06676;#mass of sphere(kg)
M0 = 1.931;


def modelFunction(x,M0,x_err,y_err):
 
    return mu/2./pi*M0/((x-x_err)**3.)+y_err

 #make residual
def residual(par,x,data=None,std=None):
 parvals=par.valuesdict()
 
 M0=parvals['M0']
 x_err=parvals['x_err']
 y_err=parvals['y_err']
 
 B = modelFunction(x,M0,x_err,y_err)
 B=B**(-1./3.)
 if data is None:
  return B
 if std is None:
  return (B-data)**2.
  
 return ((B-data)**2./std)



if __name__ == "__main__":
    
 
    text=np.loadtxt("./sphereMagnet.txt")
    x=text[:,0]
    data=text[:,1]**(-1./3.)

    #initialize Fitting parameters**************************************************
    p=lmf.Parameters()
    #           (Name,   Initial Value,     Vary,      Min  ,Max  ,Expr)
    p.add_many( ( 'M0',          1.8        ,True ,     0    , 3.  , None),
                ( 'x_err' ,      -R          ,False,    -0.013  , 0.   , None),
                ( 'y_err' ,      0.0          ,True,    -2.  , 2.   , None),
          
               )
   
    mini=lmf.Minimizer(residual,p,fcn_args=(x[xmin:xmax],data[xmin:xmax]))
    result = mini.minimize(method='leastsq')

    print(lmf.fit_report(result,show_correl=True)) #print the statistics

    plt.figure(figsize=(10,8))
    plt.plot(x,data,'x',label='Data')
    plt.plot(x,residual(result.params,x),'r-',color="brown",label='Fitting') #plot the fitting result
    plt.legend()
    plt.title('B_sphere',fontsize=16)
    plt.ylabel('B^-1/3',fontsize=12)
    plt.xlabel('x',fontsize=12)
    plt.savefig('B_sphere.png')        #save the graph
    plt.show()  

#code end

I want to know how to save & load parameters from the above fitting result. I already searched the lmfit homepage and git depository storeing some code examples. I was not able to find save & laod examples with
minimizer result.

Regards,
kooktae Kim.

Matt Newville

unread,
Jun 28, 2019, 7:24:40 AM6/28/19
to lmfit-py
--
You received this message because you are subscribed to the Google Groups "lmfit-py" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lmfit-py+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lmfit-py/eb230801-e001-42ca-b5a2-734499b0d33e%40googlegroups.com.


--
--Matt 

김국태

unread,
Jun 28, 2019, 7:43:04 AM6/28/19
to lmfi...@googlegroups.com

Oh yeah I tried it but it returns an error like this.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-10-9fcc1c47b4c2> in <module>
      1 from lmfit.model import save_modelresult
----> 2 save_modelresult(result, 'gauss_modelresult.sav')

c:\program files\python37\lib\site-packages\lmfit\model.py in save_modelresult(modelresult, fname)
   1238     """
   1239     with open(fname, 'w') as fout:
-> 1240         modelresult.dump(fout)
   1241
   1242

AttributeError: 'MinimizerResult' object has no attribute 'dump'


So I thought if I want to use residual & minimizer, I can't use save_modelresult functions from lmfit.model. 


​

​Regards, kooktae Kim 

Matt Newville

unread,
Jun 28, 2019, 8:28:31 AM6/28/19
to lmfit-py
On Fri, Jun 28, 2019 at 6:43 AM 김국태 <wksd...@naver.com> wrote:

Oh yeah I tried it but it returns an error like this.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-10-9fcc1c47b4c2> in <module>
      1 from lmfit.model import save_modelresult
----> 2 save_modelresult(result, 'gauss_modelresult.sav')

c:\program files\python37\lib\site-packages\lmfit\model.py in save_modelresult(modelresult, fname)
   1238     """
   1239     with open(fname, 'w') as fout:
-> 1240         modelresult.dump(fout)
   1241
   1242

AttributeError: 'MinimizerResult' object has no attribute 'dump'



Yes, that is correct: `save_modelresult()` saves a ModelResult, which is what is returned from `Model.fit()`, not from `minimizer`.    

So I thought if I want to use residual & minimizer, I can't use save_modelresult functions from lmfit.model. 


Yes, that is true.


`Minimizer` returns a `MinimizerResult` which is a pretty simple collection of attributes, mostly fit statistics as printed in the report, and the best-fit parameters.  The Parameters can be saved to (dump) and read from (load) a disk file.  So there really isn't much to do with a "reloaded" MinimizerResult -- you could dump the Parameters and print out the report.

--Matt

BTW, in your residual function, you probably don't want to return "(b-data)**2", just "b-data". 

김국태

unread,
Jun 28, 2019, 8:41:59 AM6/28/19
to lmfi...@googlegroups.com

Dear Matt,

 

Ok I think I solved the problem with your help.

 

I tried below code.

 

# start

import pickle


with open('par.pkl', 'wb') as f:
    pickle.dump(result, f)
   
with open('./par.pkl', 'rb') as f:
    list2 = pickle.load(f)

#end

 

It looks working well thanks!

 

Regards, kooktae Kim.

 

 

-----Original Message-----
From: "Matt Newville"<newv...@cars.uchicago.edu>
To: "lmfit-py"<lmfi...@googlegroups.com>;
Cc:
Sent: 2019-06-28 (금) 21:28:43 (GMT+09:00)
Subject: Re: saving params from minimizer result??
 

--
You received this message because you are subscribed to the Google Groups "lmfit-py" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lmfit-py+u...@googlegroups.com.

Matthew Newville

unread,
Jun 28, 2019, 8:58:41 AM6/28/19
to lmfit-py


On Fri, Jun 28, 2019, 7:41 AM 김국태 <wksd...@naver.com> wrote:

Dear Matt,

 

Ok I think I solved the problem with your help.

 

I tried below code.

 

# start

import pickle


with open('par.pkl', 'wb') as f:
    pickle.dump(result, f)
   
with open('./par.pkl', 'rb') as f:
    list2 = pickle.load(f)

#end

 

It looks working well thanks!

 


Pickle can work, but you should be aware that lmfit.Parameters is actually a pretty complex object.  If you add constraint expressions, you will run into Parameters objects that cannot be pickled.  That is why they have dump/load methods.


김국태

unread,
Jun 28, 2019, 10:28:07 PM6/28/19
to lmfi...@googlegroups.com

Dear Matt,

 

Instead using pickle, I wrote a code like this to use dumps & loads of params.

 

#start

 

def savePar( name, result )  : #get the name and result from fitting.


    par_j = result.params.dumps() # dump the results into json string.
    
    with open( name,'w' ) as json_file :
       
        json.dump( par_j, json_file ) # wirte json string into file
       
def readPar( name, result ) : #get the name and result object
    
    with open( 'p.par' , 'r' ) as json_file : #open previous result.


        json_str = json.load( json_file )


        result.params.loads( json_str ) # load previous result into input variable.
        
        return result 

 

#end

 

Does this look safe way to save and load the parameters?? (actually I wasn't able to figure out how to use dump & load. , instead I used dumps & loads that changes

fitting result into json string.) Please let me know if there is simpler & better way to do it using minimizer

 

Thanks!

 

Regards, kooktae Kim

--
You received this message because you are subscribed to the Google Groups "lmfit-py" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lmfit-py+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lmfit-py/3fad3d5e71bbf41c7ac74a7e56fb70ea%40cweb001.nm.nfra.io.
--
You received this message because you are subscribed to the Google Groups "lmfit-py" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lmfit-py+u...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages