Logo

Generic Maximum Likelihood Models

This tutorial explains how to quickly implement new maximum likelihood models in statsmodels. The GenericLikelihoodModel class eases the process by providing tools such as automatic numeric differentiation and a unified interface to scipy optimization functions. Using statsmodels, users can fit new MLE models simply by “plugging-in” a log-likelihood function.

Negative Binomial Regression for Count Data

Consider a negative binomial regression model for count data with log-likelihood (type NB-2) function expressed as:

\mathcal{L}(\beta_j; y, \alpha) = \sum_{i=1}^n y_i ln \left ( \frac{\alpha exp(X_i'\beta)}{1+\alpha exp(X_i'\beta)} \right ) - \frac{1}{\alpha} ln(1+\alpha exp(X_i'\beta)) \\ + ln \Gamma (y_i + 1/\alpha) - ln \Gamma (y_i+1) - ln \Gamma (1/\alpha)

with a matrix of regressors X, a vector of coefficients \beta, and the negative binomial heterogeneity parameter \alpha.

Using the nbinom distribution from scipy, we can write this likelihood simply as:

In [1]: import numpy as np

In [2]: from scipy.stats import nbinom

In [3]: def _ll_nb2(y, X, beta, alph):
   ...:     mu = np.exp(np.dot(X, beta))
   ...:     size = 1 / alph
   ...:     prob = size / (size + mu)
   ...:     ll = nbinom.logpmf(y, size, prob)
   ...:     return ll
   ...: 

New Model Class

We create a new model class which inherits from GenericLikelihoodModel:

In [4]: from statsmodels.base.model import GenericLikelihoodModel

In [5]: class NBin(GenericLikelihoodModel):
   ...:     def __init__(self, endog, exog, **kwds):
   ...:         super(NBin, self).__init__(endog, exog, **kwds)
   ...: 

In [6]: def nloglikeobs(self, params):
   ...:         alph = params[-1]
   ...:         beta = params[:-1]
   ...:         ll = _ll_nb2(self.endog, self.exog, beta, alph)
   ...:         return -ll
   ...: 

            # Reasonable starting values

Two important things to notice:

  • nloglikeobs: This function should return one evaluation of the negative log-likelihood function per observation in your dataset (i.e. rows of the endog/X matrix).
  • start_params: A one-dimensional array of starting values needs to be provided. The size of this array determines the number of parameters that will be used in optimization.

That’s it! You’re done!

Usage Example

The Medpar dataset is hosted in CSV format at the Rdatasets repository. We use the read_csv function from the Pandas library to load the data in memory. We then print the first few columns:

In [7]: import pandas as pd

In [8]: url = 'http://vincentarelbundock.github.com/Rdatasets/csv/COUNT/medpar.csv'

In [9]: medpar = pd.read_csv(url)

In [10]: medpar.head()
Out[10]: 
   Unnamed: 0  los  hmo  white  died  age80  type  type1  type2  type3  \
0           1    4    0      1     0      0     1      1      0      0   
1           2    9    1      1     0      0     1      1      0      0   
2           3    3    1      1     1      1     1      1      0      0   
3           4    9    0      1     0      0     1      1      0      0   
4           5    1    0      1     1      1     1      1      0      0   

   provnum  
0    30001  
1    30001  
2    30001  
3    30001  
4    30001  

The model we are interested in has a vector of non-negative integers as dependent variable (los), and 5 regressors: Intercept, type2, type3, hmo, white.

For estimation, we need to create 2 numpy arrays (pandas DataFrame should also work): a 1d array of length N to hold los values, and a N by 5 array to hold our 5 regressors. These arrays can be constructed manually or using any number of helper functions; the details matter little for our current purposes. Here, we build the arrays we need using the Patsy package:

In [11]: import patsy

In [12]: y, X = patsy.dmatrices('los~type2+type3+hmo+white', medpar)

In [13]: print y[:5]
[[ 4.]
 [ 9.]
 [ 3.]
 [ 9.]
 [ 1.]]

In [14]: print X[:5]
[[ 1.  0.  0.  0.  1.]
 [ 1.  0.  0.  1.  1.]
 [ 1.  0.  0.  1.  1.]
 [ 1.  0.  0.  0.  1.]
 [ 1.  0.  0.  0.  1.]]

Then, we fit the model and extract some information:

In [15]: mod = NBin(y, X)

In [16]: res = mod.fit()
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-16-deef2687e692> in <module>()
----> 1 res = mod.fit()

/build/statsmodels-iqIBhp/statsmodels-0.5.0+git13-g8e07d34/debian/python-statsmodels/usr/lib/python2.7/dist-packages/statsmodels/base/model.pyc in fit(self, start_params, method, maxiter, full_output, disp, callback, retall, **kwargs)
    813                             method=method, maxiter=maxiter,
    814                             full_output=full_output,
--> 815                             disp=disp, callback=callback, **kwargs)
    816         genericmlefit = GenericLikelihoodModelResults(self, mlefit)
    817 

/build/statsmodels-iqIBhp/statsmodels-0.5.0+git13-g8e07d34/debian/python-statsmodels/usr/lib/python2.7/dist-packages/statsmodels/base/model.pyc in fit(self, start_params, method, maxiter, full_output, disp, fargs, callback, retall, **kwargs)
    355                              disp=disp, maxiter=maxiter, callback=callback,
    356                              retall=retall, full_output=full_output,
--> 357                              hess=hess)
    358 
    359         if not full_output: # xopt should be None and retvals is argmin

/build/statsmodels-iqIBhp/statsmodels-0.5.0+git13-g8e07d34/debian/python-statsmodels/usr/lib/python2.7/dist-packages/statsmodels/base/model.pyc in _fit_mle_nm(f, score, start_params, fargs, kwargs, disp, maxiter, callback, retall, full_output, hess)
    481                             ftol=ftol, maxiter=maxiter, maxfun=maxfun,
    482                             full_output=full_output, disp=disp, retall=retall,
--> 483                             callback=callback)
    484     if full_output:
    485         if not retall:

/usr/lib/python2.7/dist-packages/scipy/optimize/optimize.pyc in fmin(func, x0, args, xtol, ftol, maxiter, maxfun, full_output, disp, retall, callback)
    371             'return_all': retall}
    372 
--> 373     res = _minimize_neldermead(func, x0, args, callback=callback, **opts)
    374     if full_output:
    375         retlist = res['x'], res['fun'], res['nit'], res['nfev'], res['status']

/usr/lib/python2.7/dist-packages/scipy/optimize/optimize.pyc in _minimize_neldermead(func, x0, args, callback, xtol, ftol, maxiter, maxfev, disp, return_all, **unknown_options)
    436     if retall:
    437         allvecs = [sim[0]]
--> 438     fsim[0] = func(x0)
    439     nonzdelt = 0.05
    440     zdelt = 0.00025

/usr/lib/python2.7/dist-packages/scipy/optimize/optimize.pyc in function_wrapper(*wrapper_args)
    279     def function_wrapper(*wrapper_args):
    280         ncalls[0] += 1
--> 281         return function(*(wrapper_args + args))
    282 
    283     return ncalls, function_wrapper

/build/statsmodels-iqIBhp/statsmodels-0.5.0+git13-g8e07d34/debian/python-statsmodels/usr/lib/python2.7/dist-packages/statsmodels/base/model.pyc in <lambda>(params, *args)
    327 
    328         nobs = self.endog.shape[0]
--> 329         f = lambda params, *args: -self.loglike(params, *args) / nobs
    330         score = lambda params: -self.score(params) / nobs
    331         try:

/build/statsmodels-iqIBhp/statsmodels-0.5.0+git13-g8e07d34/debian/python-statsmodels/usr/lib/python2.7/dist-packages/statsmodels/base/model.pyc in loglike(self, params)
    762 
    763     def loglike(self, params):
--> 764         return self.loglikeobs(params).sum(0)
    765 
    766     def nloglike(self, params):

/build/statsmodels-iqIBhp/statsmodels-0.5.0+git13-g8e07d34/debian/python-statsmodels/usr/lib/python2.7/dist-packages/statsmodels/base/model.pyc in loglikeobs(self, params)
    768 
    769     def loglikeobs(self, params):
--> 770         return -self.nloglikeobs(params)
    771 
    772     def score(self, params):

AttributeError: 'NBin' object has no attribute 'nloglikeobs'

Extract parameter estimates, standard errors, p-values, AIC, etc.:

In [17]: res.params
Out[17]: 
array([-0.0168,  0.0099, -0.0187, -0.0142,  0.2545,  0.2407,  0.0804,
       -1.9522, -0.3341, -0.169 ,  0.0049, -0.0036, -0.0141, -0.004 ,
       -0.0039,  0.0917,  0.049 ,  0.008 ,  0.0002, -0.0022,  2.9589])

In [18]: res.bse
Out[18]: 
array([ 0.0004,  0.0006,  0.0007,  0.0004,  0.0299,  0.0571,  0.0139,
        0.3168,  0.0613,  0.0327,  0.0013,  0.0002,  0.0019,  0.0005,
        0.001 ,  0.0145,  0.0075,  0.0015,  0.    ,  0.0003,  1.5467])

In [19]: res.pvalues
Out[19]: 
array([ 0.    ,  0.    ,  0.    ,  0.    ,  0.    ,  0.    ,  0.    ,
        0.    ,  0.    ,  0.    ,  0.0001,  0.    ,  0.    ,  0.    ,
        0.    ,  0.    ,  0.    ,  0.    ,  0.    ,  0.    ,  0.0557])

In [20]: res.aic
Out[20]: 6039.2251179879531

As usual, you can obtain a full list of available information by typing dir(res).

To ensure that the above results are sound, we compare them to results obtained using the MASS implementation for R:

url = 'http://vincentarelbundock.github.com/Rdatasets/csv/COUNT/medpar.csv'
medpar = read.csv(url)
f = los~factor(type)+hmo+white

library(MASS)
mod = glm.nb(f, medpar)
coef(summary(mod))
                 Estimate Std. Error   z value      Pr(>|z|)
(Intercept)    2.31027893 0.06744676 34.253370 3.885556e-257
factor(type)2  0.22124898 0.05045746  4.384861  1.160597e-05
factor(type)3  0.70615882 0.07599849  9.291748  1.517751e-20
hmo           -0.06795522 0.05321375 -1.277024  2.015939e-01
white         -0.12906544 0.06836272 -1.887951  5.903257e-02

Numerical precision

The statsmodels and R parameter estimates agree up to the fourth decimal. The standard errors, however, agree only up to the second decimal. This discrepancy may be the result of imprecision in our Hessian numerical estimates. In the current context, the difference between MASS and statsmodels standard error estimates is substantively irrelevant, but it highlights the fact that users who need very precise estimates may not always want to rely on default settings when using numerical derivatives. In such cases, it may be better to use analytical derivatives with the LikelihoodModel class.

Table Of Contents

This Page