statsmodels: AttributeError: module 'scipy.stats' has no attribute 'chisqprob'

Error: c:\program files (x86)\python35-64\lib\site-packages\statsmodels\discrete\discrete_model.py in summary(self, yname, xname, title, alpha, yname_list) 2756 yname_list=None): 2757 smry = super(BinaryResults, self).summary(yname, xname, title, alpha, -> 2758 yname_list) 2759 fittedvalues = self.model.cdf(self.fittedvalues) 2760 absprederror = np.abs(self.model.endog - fittedvalues)

c:\program files (x86)\python35-64\lib\site-packages\statsmodels\discrete\discrete_model.py in summary(self, yname, xname, title, alpha, yname_list) 2548 (‘Log-Likelihood:’, None), 2549 (‘LL-Null:’, [“%#8.5g” % self.llnull]), -> 2550 (‘LLR p-value:’, [“%#6.4g” % self.llr_pvalue]) 2551 ] 2552

c:\program files (x86)\python35-64\lib\site-packages\statsmodels\tools\decorators.py in get(self, obj, type) 95 if _cachedval is None: 96 # Call the “fget” function —> 97 _cachedval = self.fget(obj) 98 # Set the attribute in obj 99 # print(“Setting %s in cache to %s” % (name, _cachedval))

c:\program files (x86)\python35-64\lib\site-packages\statsmodels\discrete\discrete_model.py in llr_pvalue(self) 2403 @cache_readonly 2404 def llr_pvalue(self): -> 2405 return stats.chisqprob(self.llr, self.df_model) 2406 2407 @cache_readonly

AttributeError: module ‘scipy.stats’ has no attribute ‘chisqprob’

To reproduce the error: Dataset information - http://www.statsmodels.org/dev/datasets/generated/ccard.html

Code:

df = sm.datasets.ccard.load_pandas().data
df['intercept'] = 1
df = df[['intercept', 'AGE', 'INCOME','OWNRENT']]
model = sm.Logit(df.OWNRENT, df[['intercept', 'AGE', 'INCOME']])
result = model.fit()
result.summary()

Issue: File - statsmodels\discrete\discrete_model.py Function - llr_pvalue(self)

This returns - stats.chisqprob(self.llr, self.df_model) But the function stats.chisqprob is depricated as of statsmodel 1.7 (https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.chisqprob.html) and should be replaced with stats.distributions.chi2.sf

Ouput of sm.show_versions()

INSTALLED VERSIONS
------------------
Python: 3.5.3.final.0

Statsmodels
===========

Installed: 0.8.0 (c:\program files (x86)\python35-64\lib\site-packages\statsmodels)

Required Dependencies
=====================

cython: Not installed
numpy: 1.13.1 (c:\program files (x86)\python35-64\lib\site-packages\numpy)
scipy: 1.0.0b1 (c:\program files (x86)\python35-64\lib\site-packages\scipy)
pandas: 0.20.3 (c:\program files (x86)\python35-64\lib\site-packages\pandas)
    dateutil: 2.6.1 (c:\program files (x86)\python35-64\lib\site-packages\dateutil)
patsy: 0.4.1 (c:\program files (x86)\python35-64\lib\site-packages\patsy

About this issue

Original URL
State: closed
Created 7 years ago
Comments: 16 (3 by maintainers)

Links to this issue

python - statsmodel AttributeError: module 'scipy.stats' has no attribute 'chisqprob' - Stack Overflow

Commits related to this issue

BUG: Remove deprecated scipy function Remove deprecated chisqprob in favor of stats.distributions.chi2.sf closes #3931 — committed to bashtage/statsmodels by bashtage 7 years ago
BUG: Remove deprecated scipy function Remove deprecated chisqprob in favor of stats.distributions.chi2.sf closes #3931 — committed to tupui/statsmodels by bashtage 7 years ago

Most upvoted comments

Running into this issue myself now. In case it wasn’t immediately obvious to anyone else, the workaround for this specific issue with missing function chisqprob is

from scipy import stats
stats.chisqprob = lambda chisq, df: stats.chi2.sf(chisq, df)

+76

VincentLa14 on Nov 13, 2017

@NaderNazemi This will not be fixed (reverted) in scipy (given also the existence of a simple workaround). It is our fault that we missed this and that we are slow with a new release.

However, there is a simple workaround by assigning the missing function back into the scipy.stats namespace as shown above, i.e.

from scipy import stats
stats.chisqprob = lambda chisq, df: stats.chi2.sf(chisq, df)

+15

josef-pkt on Feb 13, 2018

results.summary2() will work

Model:	Logit	No. Iterations:	6.0000
Dependent Variable:	converted	Pseudo R-squared:	0.000
Date:	2018-05-29 13:35	AIC:	215704.9004
No. Observations:	294478	BIC:	215726.0864
Df Model:	1	Log-Likelihood:	-1.0785e+05
Df Residuals:	294476	LL-Null:	-1.0785e+05
Converged:	1.0000	Scale:	1.0000

mynamerahulkumar on May 29, 2018

instead of downgrading scipy, it’s better to upgrade statsmodels. statsmodels 0.9 has the compatibility fixes

josef-pkt on Jul 3, 2018

Thanks for checking and reporting. I didn’t know we still use those functions. Based on a file search it is the only use of chisqprob, but there are a few more removed distribution function.

Note this is in DiscreteResults so it will cause errors in all discrete models, i.e all summary will be “dead”.

@srivathsadv Did you run the statsmodels test suite with scipy 1.0 candidate? Are there other errors?

workaround is to assign the missing function to scipy.stats

>>> from scipy import stats
>>> stats.junk = lambda chisq, df: stats.chi2.sf(chisq, df)
>>> stats.junk(10, 3)
>>> 0.018566135463043251
>>> stats.chisqprob(10,3)
>>> 0.018566135463043251

josef-pkt on Sep 25, 2017