函数scipy.optimize.fmin for python

Question

I'm trying to minimize a khi-square using scipy.optimize.fmin. Here is my function, (which calls an other simulation function spotdiffusion). The returned value (chi) is an array of two khi values (one for congruent condition, the other for incongruent condition) which I try to minimize:

def chis (a, ter , v , sda , rd):

    ncond=1 
    ntrials = 1000 
    observed_data = np.array ([ [0.9995835, 24.0, 329.5, 357.9, 370.5, 391.5, 457.6, 0.0004164931, 0, 0],#congruent cond
                                [0.6953498, 16,   409.5, 450.5, 481,   529,   546 ,  0.3046502 ,  7 ,350]])#incongruent cond

    q_probs=np.array ([.1,.2,.2,.2,.2,.1])   
    b_probs=np.array([0.501,0.499])   

    cond = np.arange (0, ncond)
    chi = []
    for g in cond:
        if(g==0):
            fl= 1.0   #flankers congruent with target    
        if(g==1):
            fl= -1.0   # incongruent

        #########
        simTRcorrect, simTRerror, simprobc, simprobe = spotdiffusion (a ,ter ,v, sda,rd ,fl, ntrials = 1000)
        #########        
        top_data = observed_data[g,0]*q_probs  
        bot_data=observed_data[g,7]*b_probs    

        pt1 = (len (simTRcorrect [simTRcorrect < observed_data[g, 2]])) /ntrials 
        pt2 = (len (simTRcorrect [(simTRcorrect < observed_data[g, 3]) & (simTRcorrect >= observed_data[g, 2])])) /ntrials
        pt3 = (len (simTRcorrect [(simTRcorrect < observed_data[g, 4]) & (simTRcorrect >= observed_data[g, 3])])) /ntrials 
        pt4 = (len (simTRcorrect [(simTRcorrect < observed_data[g, 5]) & (simTRcorrect >= observed_data[g, 4])])) /ntrials
        pt5 = (len (simTRcorrect [(simTRcorrect < observed_data[g, 6]) & (simTRcorrect >= observed_data[g, 5])])) /ntrials
        pt6=(len (simTRcorrect [simTRcorrect > observed_data[g, 6]])) /ntrials

        pred_p= np.array ([pt1,pt2,pt3,pt4,pt5,pt6])
        top_chi_array = (np.square (top_data-pred_p))/ (pred_p+ 0.001)
        top_chi = np.sum (top_chi_array) 


        pt1 = (len (simTRerror[simTRerror < observed_data[g, 9]]))  /ntrials
        pt2 = (len (simTRerror[simTRerror >= observed_data[g, 9]])) /ntrials

        pred_p=np.array ([pt1,pt2])
        bot_chi_array = (np.square (bot_data-pred_p)) / (pred_p+ 0.001)
        bot_chi= np.sum (bot_chi_array)

        totchi=(bot_chi+top_chi)*(observed_data[g,1]+ observed_data[g,8])


        chi.append (totchi)


    chi = np.array (chi)       
    return chi

Here is the fitting procedure:

x0 = np.array ([0.11, 0.25,0.35,1.7,0.017]) ####for initial guess 
xopt = fmin (chis(a, ter , v , sda , rd), x0, maxiter=300)

I've got an error that I don't understand:

Traceback (most recent call last):
  File "<ipython console>", line 1, in <module>
  File "C:\Python27\lib\site-packages\spyderlib\widgets\externalshell\startup.py", line 128, in runfile
    execfile(filename, glbs)
  File "C:\Users\mathieu\Desktop\modeling\spotlight diffusion model\fitting_spotlight.py", line 245, in <module>
    xopt = fmin (chis(a, ter , v , sda , rd), x0, maxiter=300)
  File "C:\Python27\lib\site-packages\scipy\optimize\optimize.py", line 257, in fmin
    fsim[0] = func(x0)
  File "C:\Python27\lib\site-packages\scipy\optimize\optimize.py", line 176, in function_wrapper
    return function(x, *args)
TypeError: 'numpy.float64' object is not callable

Does anyone have an idea of what's going wrong?

Cheers, Mat

Answer 1

The problem is in this line:

xopt = fmin (chis(a, ter , v , sda , rd), x0, maxiter=300)

The expression

chis(a, ter , v , sda , rd)

is most likely number. It is the result of calling the function chis .

Instead, we want to pass the function object chis to the fmin function, without having called chis first. (If we pass chis(a, ter, v, sda, rd) then fmin just gets a number as its first argument. If we pass the function object chis itself, then fmin can call chis how ever it needs to from within the body of fmin . In Python, functions are first-class objects.

So try instead

xopt = fmin (chis, x0, maxiter=300)

Answer 2

the problem seems to be both - in line:

xopt=fmin(chis(a,ter,v,sda,rd),x0,maxiter=300)

which should be as previous user mentioned

xopt=fmin(chis,x0,maxiter=300)

but also in the beginning, where function has been defined, parameters should be given as array

instead of

def chis (a, ter , v , sda , rd):

try this:

def chis (arrays):

    a=arrays[0]
    ter=arrays[1]
    v=arrays[2]
    sda=arrays[3]
    rd=arrays[4]