Statsmodels: Pickling ARIMA fails

[ksomemo]

ARIMA class includes dates in __getnewargs__ method. but model do not includes dates . my version is '0.8.0'.

```code.py
from statsmodels.compat.python import cPickle

cPickle.dumps(arima_mod)

=> AttributeError: 'ARIMA' object has no attribute 'dates'

type(arima_mod)

=> statsmodels.tsa.arima_model.ARIMA

```arima_model.py
    def __getnewargs__(self):
        return ((self.endog),
                (self.k_lags, self.k_diff, self.k_ma),
                self.exog, self.dates, self.freq, self.missing)

test code mod is actually ARMA model. at statsmodels/tsa/tests/test_arima.py.

def test_arima_pickle():
    endog = y_arma[:, 6]
    mod = ARIMA(endog, (1, 0, 1)

[josef-pkt]

Sorry, I'm slow in catching up with issues that are not on my current topic list.

to clarify the issue

if we use ARIMA(1, 1, 1) in test_arima_pickle, then it raises an error about missing dates

======================================================================
ERROR: statsmodels.tsa.tests.test_arima.test_arima_pickle
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Users\josef\Downloads\WinPython-64bit-3.4.4.5Qt5\python-3.4.4.amd64\l
ib\site-packages\nose\case.py", line 198, in runTest
    self.test(*self.arg)
  File "E:\Josef_new\eclipsews\statsmodels_py34_pr\statsmodels\tsa\tests\test_ar
ima.py", line 2276, in test_arima_pickle
    pkl_mod = cPickle.loads(cPickle.dumps(mod))
  File "E:\Josef_new\eclipsews\statsmodels_py34_pr\statsmodels\tsa\arima_model.p
y", line 994, in __getnewargs__
    self.exog, self.dates, self.freq, self.missing)
AttributeError: 'ARIMA' object has no attribute 'dates'

[josef-pkt]

using the debugger, in the test example the other keywords are also not available except for exog

(Pdb) self.dates
*** AttributeError: 'ARIMA' object has no attribute 'dates'
(Pdb) self.freq
*** AttributeError: 'ARIMA' object has no attribute 'freq'
(Pdb) self.exog
(Pdb) self.missing
*** AttributeError: 'ARIMA' object has no attribute 'missing'

[josef-pkt]

@ksomemo Thank you for reporting

PR with fix is in #3785

as reference: previous PRs for __getnewargs__ #3217 and #3412

[cristofp]

Any known workaround before new version of library is released?

[josef-pkt]

There is not direct workaround except for using the usual Python monkey patching or flexibility
e.g. setting the missing attributes might/should work
model.dates = None
model.freq = None
model.missing = None

(I haven't tried it.)