AttributeError: 'numpy.timedelta64' object has no attribute 'days' about prophet

[Weiss-Robin]

Had the same issue when using a virtualenv with Python 2.7. Creating a new one with Python 3.5.1 solved the problem for me.

[zarak]

Here's my version information and the full traceback:

Ubuntu 16.04.1 LTS
Python 3.5.2 |Anaconda 2.4.1 (64-bit)| (default, Jul 2 2016, 17:53:06)

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-8-a76c855f084a> in <module>()
----> 1 m.fit(df)

/home/z/anaconda3/lib/python3.5/site-packages/fbprophet/forecaster.py in fit(self, df)
    337         history = self.setup_dataframe(history, initialize_scales=True)
    338         self.history = history
--> 339         seasonal_features = self.make_all_seasonality_features(history)
    340 
    341         self.set_changepoints()

/home/z/anaconda3/lib/python3.5/site-packages/fbprophet/forecaster.py in make_all_seasonality_features(self, df)
    286                 365.25,
    287                 10,
--> 288                 'yearly',
    289             ))
    290 

/home/z/anaconda3/lib/python3.5/site-packages/fbprophet/forecaster.py in make_seasonality_features(cls, dates, period, series_order, prefix)
    227     @classmethod
    228     def make_seasonality_features(cls, dates, period, series_order, prefix):
--> 229         features = cls.fourier_series(dates, period, series_order)
    230         columns = [
    231             '{}_{}'.format(prefix, i + 1)

/home/z/anaconda3/lib/python3.5/site-packages/fbprophet/forecaster.py in fourier_series(dates, period, series_order)
    216         t = np.array(
    217             (dates - pd.datetime(1970, 1, 1))
--> 218             .apply(lambda x: x.days)
    219             .astype(np.float)
    220         )

/home/z/anaconda3/lib/python3.5/site-packages/pandas/core/series.py in apply(self, func, convert_dtype, args, **kwds)
   2167             values = lib.map_infer(values, lib.Timestamp)
   2168 
-> 2169         mapped = lib.map_infer(values, f, convert=convert_dtype)
   2170         if len(mapped) and isinstance(mapped[0], Series):
   2171             from pandas.core.frame import DataFrame

pandas/src/inference.pyx in pandas.lib.map_infer (pandas/lib.c:62578)()

/home/z/anaconda3/lib/python3.5/site-packages/fbprophet/forecaster.py in <lambda>(x)
    216         t = np.array(
    217             (dates - pd.datetime(1970, 1, 1))
--> 218             .apply(lambda x: x.days)
    219             .astype(np.float)
    220         )

AttributeError: 'numpy.timedelta64' object has no attribute 'days'

I didn't have any issues running this with conda on OSx.

[zarak]

The approaches described in the pandas docs both work for me, but I'm still not sure why it works on my Mac but not on my Linux system.

# to days
In [70]: td / np.timedelta64(1,'D')
Out[70]: 
0    31.000000
1    31.000000
2    31.003507
3          NaN
dtype: float64

In [71]: td.astype('timedelta64[D]')
Out[71]: 
0    31
1    31
2    31
3   NaN
dtype: float64

[zarak]

@robin-weiss I think you're right. I upgraded to Anaconda2-4.3.0 and it works fine now. The problem seems to be because of an older version of pandas.

[shengbaodong]

I got the same problem on my mac with python2.7.10 + pandas-0.17.1, use pip install -U pandas ; I update my pandas to 0.19.2, and the problem solved