0

I am trying to minimize a simple GMM criterion, calling variables from a Pandas dataframe. Each column name present a column in the dataframe (492 observations). In the first block I define matrices and vector from the dataframe, in the second part I compute a method of moments criterion that should be minimized by setting the right coefficients coeffs. The objects are conformable for matrix multiplication, and if I do it step by step using the initial value for coeff I finally get a scalar. However, if it does not work when I pass the function to the minimizer. 

import numpy             as np
import pandas            as pd

from numpy.linalg   import inv
from scipy.optimize import fsolve, minimize, fmin_bfgs

df      = pd.read_csv('data.csv')

dep   = ['dep']                      # A column name in my dataframe
exog  = ['exog1', 'exog2', 'exog3']  # A list of columns
endog = ['endog']
instr = ['instr']
coeff = np.array([[0,0,0,0]]).T


def GMM_criterion(coeff, df, dep, exog, endog, instr):

    # Set up variables
    x     = endog + exog 
    z     = exog  + instr 
    n     = df.shape[0]
    x     = df[x].values    # A (492 x 4) matrix
    z     = df[z].values    # A (492 x 4) matrix
    y     = df[dep].values  # A (492 x 1) vector
    y_hat = x @ coeff       # A (492 x 1) vector of predicted values
    resid = y - y_hat       # A (492 x 1) vector of residuals

# Compute GMM criterion
    W     =  1/n * inv(z.T @ z)  # A (4x4) weighting matrix
    g     =  1/n * z.T @ resid   # A (4x1) vector
    GMM   = n * g.T @ W @ g      # Becomes a (1x1) numpy.ndarray
    GMM   = GMM.flatten()
    GMM   = float(GMM)

    return GMM

GMM = minimize(GMM_criterion, coeff, args=(df, dep, exog, endog, instr), 
method='SLSQP')

I get this error only size-1 arrays can be converted to Python scalars.

Traceback (most recent call last):

File "<ipython-input-300-8809d7df9a20>", line 31, in <module>
TSLS = minimize(GMM_criterion, coeff, args=(df, dep, exog, endog, instr), 
method='SLSQP')

File "C:\Users\Igor\Anaconda3\lib\site- 
packages\scipy\optimize\_minimize.py", line 611, in minimize
constraints, callback=callback, **options)

File "C:\Users\Igor\Anaconda3\lib\site-packages\scipy\optimize\slsqp.py", 
line 379, in _minimize_slsqp
fx = func(x)

File "C:\Users\Igor\Anaconda3\lib\site- 
packages\scipy\optimize\optimize.py", 
line 293, in function_wrapper
return function(*(wrapper_args + args))

File "<ipython-input-300-8809d7df9a20>", line 25, in GMM_criterion
GMM   = float(GMM)

TypeError: only size-1 arrays can be converted to Python scalars`

Do you have any suggestion on how to fix this? Thanks

Nick ODell
  • 15,465
  • 3
  • 32
  • 66
iiiccciii
  • 1
  • 1
  • Please fix your indentation, define all imports, variables, and inputs used, and tell us the expected output. In short, provide a [Minimal, Complete, and Verifiable example](https://stackoverflow.com/help/mcve). This includes a full traceback of the error messages. – Mr. T Oct 27 '18 at 05:49
  • Thanks, I tried to fix the issues and provided more details. – iiiccciii Oct 27 '18 at 16:02
  • According to the error message, GMM is probably an array with a length of at least two. Python/Numpy is user-friendly and tries to convert arrays of length 1, e.g., `float(np.array([4]))`. But what should be `float(np.array([2, 3, 4]))`? – Mr. T Oct 27 '18 at 16:37
  • The dimension of GMM is 1x1, if I try doing it with some random coefficients it works. The problem arises when I use it as a function – iiiccciii Oct 27 '18 at 18:02

0 Answers0