predict yields array of 0s.

[zachlefevre]

Question

data = [359125698.64, -419934.7, -31623185.37, 19556515.5, 934374.83, 88643847.73, 18983611.85, 638945.66, -4079020.52, -41301474.38, -999558.35, -999558.35, 52100319.7, -784917.62, 48527345.98, -3161926.01, 1757748.53, 86546250.69, 66971847.0, 1386714.85, -13195016.89, -19012979.5, 1152601.3, 11968257.33, 71107580.44, -1156768.56, -27878666.6, 888604.13, -29190832.71, 5499733.88, -23377004.53, -91490448.09, 7413504.15, 19425919.04, 742786.38, -3862882.17]

auto_arima(data, seasonal=False).predict() # yields [0., 0., ...]. order=(0, 0, 0)

Why does this only produce 0s? statsmodels.tsa.arima_model.ARIMA(data, order=(0,0,0)).fit().predict() does not.

Am I able to pull a statsmodel ARIMA from the AutoArima object?

Version:
System:
python: 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 13:42:17) [GCC 4.2.1 Compatible Clang 4.0.1 (tags/RELEASE_401/final)]
executable: /Users/zacharylefevre/anaconda3/envs/forecasting-service/bin/python
machine: Darwin-19.5.0-x86_64-i386-64bit

Python dependencies:
pip: 20.0.2
setuptools: 46.4.0.post20200518
sklearn: 0.23.1
statsmodels: 0.11.1
numpy: 1.18.1
scipy: 1.5.0
Cython: 0.29.17
pandas: 1.0.3
joblib: 0.16.0
pmdarima: 1.6.1

[aaronreidsmith]

statsmodels.tsa.arima_model.ARIMA has been all but deprecated by statsmodels (see this comment and corresponding PR in our project), so we use statsmodels.tsa.statespace.sarimax.SARIMAX under the hood.

Am I able to pull a statsmodel ARIMA from the AutoArima object?

Yes, you can access this through the arima_res_ attribute of the returned model

With all of that being said, it should be noted that statsmodels.tsa.statespace.sarimax.SARIMAX returns the same predictions in this case:

>>> statsmodels.tsa.statespace.sarimax.SARIMAX(data, order=(0,0,0)).fit().predict()
RUNNING THE L-BFGS-B CODE

           * * *

Machine precision = 2.220D-16
 N =            1     M =           10
 This problem is unconstrained.

At X0         0 variables are exactly at the bounds

At iterate    0    f=  1.94764D+01    |proj g|=  3.55318D-10

           * * *

Tit   = total number of iterations
Tnf   = total number of function evaluations
Tnint = total number of segments explored during Cauchy searches
Skip  = number of BFGS updates skipped
Nact  = number of active bounds at final generalized Cauchy point
Projg = norm of the final projected gradient
F     = final function value

           * * *

   N    Tit     Tnf  Tnint  Skip  Nact     Projg        F
    1      0      1      0     0     0   3.553D-10   1.948D+01
  F =   19.476442963321666     

CONVERGENCE: NORM_OF_PROJECTED_GRADIENT_<=_PGTOL            
array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0.])