statsmodels: how to add a method to a class of MLE result

Question

(This is not a problem of simply adding a method to a given class)

What I Want to Achieve

Using Maximum Likelihood Estimation (Generic models) of statsmodels, I implemented an MLE estimator, and want to add a user-made method, which uses exog and params, to a class of fitted result (not an instance), e.g., using classmetod(). But an error occurs because those variables are not available. How can I achieve my goal?

Let me explain what I have done so far, using an example from here.

(I had a look at this for adding a method to an existing class.)

Example

import numpy as np
from scipy import stats
import statsmodels.api as sm
from statsmodels.base.model import GenericLikelihoodModel,GenericLikelihoodModelResults

data = sm.datasets.spector.load_pandas()
endog = data.endog
exog = sm.add_constant(data.exog)

class MyProbit(GenericLikelihoodModel):
    def loglike(self, params):
        exog = self.exog
        endog = self.endog
        q = 2 * endog - 1
        return stats.norm.logcdf(q*np.dot(exog, params)).sum()

# my attemp starts ---------------
def my_method(self):
    return print(self.exog, self.params, self.model)

GenericLikelihoodModelResults.my_method = classmethod(my_method)
# my attemp ends ----------------

res = MyProbit(endog, exog).fit()

res.my_method()

This generates the following error.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-29-a2d4f516bca7> in <module>
     23 res = MyProbit(endog, exog).fit()
     24 
---> 25 res.my_method()

<ipython-input-29-a2d4f516bca7> in my_method(self)
     17 # my attemp start ---------------
     18 def my_method(self):
---> 19     return print(self.exog, self.params, self.model)
     20 GenericLikelihoodModelResults.my_method = classmethod(my_method)
     21 # my attemp ends ----------------

AttributeError: type object 'GenericLikelihoodModelResults' has no attribute 'exog'

This suggests that exog (similarly, endog and params) are not available in GenericLikelihoodModelResults. Indeed, adding the following code shows none of exog, etc.

def my_check(self):
    return dir(self)

GenericLikelihoodModelResults.my_check = classmethod(my_check)

This is despite the fact that they are available at an instance, as one can check using

res.exog
res.endog
res.params

I appreciate any constructive suggestions/comments.

score 0 · Answer 1 · answered Jun 24 '19 at 05:00

The exception message

AttributeError: type object 'GenericLikelihoodModelResults' has no attribute 'exog'

refers to the results class returned by fit.
exog, endog and similar are attributes of the model class. But params is a results attribute because it is an outcome of the fit method

So either access the model attribute of in the results class self.model.exog if self is a results instance, or use the method for the model class, i.e. when self is a MyProbit instance.

I'm not sure about the use of a classmethod in this case. The attributes are only available for the specific instance of the class.

Using `my_check()` function in my question, I could not find `model` in the result class. No chance in the model class, either. Thanks for your comment, anyway. — T_T, Jun 24 '19 at 23:28

statsmodels: how to add a method to a class of MLE result

1 Answers1