ValueWarning: No frequency information was provided, so inferred frequency MS will be used

Question

I try to fit Autoregression by sm.tsa.statespace.SARIMAX. But I meet a warning, then I want to set frequency information for this model. Who used to meet it, can you help me ?

fit1 = sm.tsa.statespace.SARIMAX(train.Demand, order=(1, 0, 0), 
                            enforce_stationarity=False,
                            enforce_invertibility=False).fit()
y_hat['AR'] = fit1.predict(start="1975-01-01", end="1975-12-01", dynamic=True)
plt.figure(figsize=(16,8))
plt.plot( train['Demand'], label='Train')
plt.plot(test['Demand'], label='Test')
plt.plot(y_hat_avg['AR'], label='AR')
plt.legend(loc='best')
plt.show()

C:\Users\thach.le\Anaconda3\lib\site-packages\statsmodels-0.8.0-py3.6-win- 
amd64.egg\statsmodels\tsa\base\tsa_model.py:165: ValueWarning: No frequency 
information was provided, so inferred frequency MS will be used.
% freq, ValueWarning)

Thanks

Dedededen

This question could also be tagged with [pandas], since the error only arises when sending a datetime-indexed pandas DataFrame to the statsmodels function. – Lorient

Just shut it down from statsmodels.tools.sm_exceptions import ValueWarning warnings.simplefilter('ignore', ValueWarning) – Puma

Answer 1

If your data is really periodic and you don't have gaps in your time series, then pandas can infer the frequency.

If the inferred frequency looks correct to you, you can use it by follow the answer at Set pandas.tseries.index.DatetimeIndex.freq with inferred_freq

For example

train.index = pd.DatetimeIndex(train.index.values,
                               freq=train.index.inferred_freq)
fit1 = sm.tsa.statespace.SARIMAX(...)

But note that this can still give a DatetimeIndex whose frequency is None if your data is not truly periodic.

For example, if you have daily data and one day is missing, then the inferred_freq will be None and attempting to pass freq="D" will raise a ValueError exception. In this case, try building your DataFrame so that all dates are present and the values in the column(s) you're forecasting are None on those dates. Then you can use missing="drop" (or whatever) with your ARIMA model.

Lorient

Is there a reason not to do train.index.freq = train.index.inferred_freq? - Never mind, found it in the linked question. – Elector