Additional library support for SARIMAX models from darts package?

[akshobhya95]

Hi , I was looking into this github and I see that I am also facing similar issue over darts package to support for SARIMA and SARIMAX models to use it https://github.com/unit8co/darts/issues/854 ?

I tried installing u8darts[all] from the requirements.txt but I don't find these models to be specific. Kindly showcase the path if it already exists? else is there a similar workaround to use specific to SARIMAX models. I don't need to use ARIMA for my case.

-Kind Regards , Akshobhya.

[madtoinou]

Hi @akshobhya95,

The closing message from the linked issue is saying that the Darts ARIMA model is a wrapper around statsmodel ARIMA, which according to their own documentation, is an interface to the ARIMA-based models (including SARIMA) : here. However, Darts also adds exogenous variables support (covariates), hence making it equivalent to the SARIMAX model. The only difference is that some of SARIMAX' arguments such as time_varying_regression, mle_regression, simple_differencing or hamilton_representation are not accessible.

Does it answer your question?

[akshobhya95]

Hi @madtoinou , I tried with the above suggestion as mentioned over the github link, but especially with ARIMA model test I am hitting this LU decomposition error in the middle of generating forecasts specifically hitting with ARIMA models for my dataset whereas for the rest of the models it is working fine.

Generating forecasts... C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals warnings.warn("Maximum Likelihood optimization failed to " 19%|█████████████████████████▏ | 243/1266 [04:13<17:48, 1.04s/it] Traceback (most recent call last): File "C:\Users\aksho\anaconda3\conformal-time-series\tests\base_test.py", line 86, in data['forecasts'] = generate_forecasts(data, *args['sequences'][0]) File "C:\Users\aksho\anaconda3\conformal-time-series\tests..\core\model_scores.py", line 90, in generate_forecasts model_forecasts = model.historical_forecasts(y,future_covariates=exog,forecast_horizon=fit_every, retrain=retrain, verbose=True).values()[:,0].squeeze() File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\darts\utils\utils.py", line 143, in sanitized_method return method_to_sanitize(self, only_args.values(), only_kwargs) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\darts\models\forecasting\forecasting_model.py", line 1051, in historical_forecasts model._fit_wrapper( File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\darts\models\forecasting\forecasting_model.py", line 386, in _fit_wrapper self.fit(series=series, add_kwargs, kwargs) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\darts\models\forecasting\forecasting_model.py", line 2788, in fit return self._fit(series, future_covariates=future_covariates) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\darts\models\forecasting\arima.py", line 167, in _fit self.model = m.fit() File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\tsa\arima\model.py", line 395, in fit res = super().fit( File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\tsa\statespace\mlemodel.py", line 703, in fit mlefit = super().fit(start_params, method=method, File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\base\model.py", line 566, in fit xopt, retvals, optim_settings = optimizer._fit(f, score, start_params, File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\base\optimizer.py", line 243, in _fit xopt, retvals = func(objective, gradient, start_params, fargs, kwargs, File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\base\optimizer.py", line 660, in _fit_lbfgs retvals = optimize.fmin_l_bfgs_b(func, start_params, maxiter=maxiter, File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\scipy\optimize_lbfgsb_py.py", line 237, in fmin_l_bfgs_b res = _minimize_lbfgsb(fun, x0, args=args, jac=jac, bounds=bounds, File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\scipy\optimize_lbfgsb_py.py", line 407, in _minimize_lbfgsb f, g = func_and_grad(x) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\scipy\optimize_differentiable_functions.py", line 296, in fun_and_grad self._update_fun() File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\scipy\optimize_differentiable_functions.py", line 262, in _update_fun self._update_fun_impl() File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\scipy\optimize_differentiable_functions.py", line 163, in update_fun self.f = fun_wrapped(self.x) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\scipy\optimize_differentiable_functions.py", line 145, in fun_wrapped fx = fun(np.copy(x), args) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\base\model.py", line 534, in f return -self.loglike(params, args) / nobs File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\tsa\statespace\mlemodel.py", line 938, in loglike loglike = self.ssm.loglike(complex_step=complex_step, kwargs) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\tsa\statespace\kalman_filter.py", line 1001, in loglike kfilter = self._filter(**kwargs) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\tsa\statespace\kalman_filter.py", line 921, in _filter self._initialize_state(prefix=prefix, complex_step=complex_step) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\statsmodels\tsa\statespace\representation.py", line 1058, in _initialize_state self._statespaces[prefix].initialize(self.initialization, File "statsmodels\tsa\statespace\_representation.pyx", line 1373, in statsmodels.tsa.statespace._representation.dStatespace.initialize File "statsmodels\tsa\statespace\_representation.pyx", line 1362, in statsmodels.tsa.statespace._representation.dStatespace.initialize File "statsmodels\tsa\statespace\_initialization.pyx", line 288, in statsmodels.tsa.statespace._initialization.dInitialization.initialize File "statsmodels\tsa\statespace\_initialization.pyx", line 406, in statsmodels.tsa.statespace._initialization.dInitialization.initialize_stationary_stationary_cov File "statsmodels\tsa\statespace\_tools.pyx", line 1548, in statsmodels.tsa.statespace._tools._dsolve_discrete_lyapunov numpy.linalg.LinAlgError: LU decomposition error.

But I get your point with the above explanation makes sense to me it answers my question. Thanks for your response.

-Kind Regards , Akshobhya.

[akshobhya95]

Hi @madtoinou , My bad I accidentally submitted my update with closing comments, Still I had one more query about the DARTS package used in ARIMA, as you said DARTS in ARIMA supports only fewer hyper tuning parameters for my case I was hitting this LU decomposition error as I have highlighted in above update, In one of the workarounds I need to hyper tune it by setting and check enforce_stationarity= False ARIMA But unfortunately I don't see this hyperparameter being used in DARTS ? is my understanding right that DARTS in ARIMA only supports fewer hypertuning parameters ? else How do we set this explicitly using DARTS?

Generating forecasts... Traceback (most recent call last): File "C:\Users\aksho\anaconda3\conformal-time-series\tests\base_test.py", line 86, in data['forecasts'] = generate_forecasts(data, args['sequences'][0]) File "C:\Users\aksho\anaconda3\conformal-time-series\tests..\core\model_scores.py", line 71, in generate_forecasts model = ARIMA(p=3,d=0,q=0,seasonal_order=(1,1,1,7),enforce_stationarity=False) File "C:\Users\aksho\anaconda3\envs\qls\lib\site-packages\darts\models\forecasting\forecasting_model.py", line 114, in call return super().call(all_params) TypeError: ARIMA.init() got an unexpected keyword argument 'enforce_stationarity'

I see that the enforce_stationarity argument is also involved in the case so kindly confirm on this check.

[madtoinou]

Hi @akshobhya95, Darts is indeed no giving access to these arguments and I added it to the roadmap.

In the meantime, you could try to make your series stationarity so that you don't have to change the enforce_stationarity argument.

[akshobhya95]

Sure Thanks for your Confirmation on this @madtoinou.