Merge branch 'master' into feature/rnn-normalization

darts/models/forecasting/arima.py

@@ -32,7 +32,7 @@ def __init__(
         q: int = 0,
         seasonal_order: Tuple[int, int, int, int] = (0, 0, 0, 0),
         trend: Optional[str] = None,
-        random_state: int = 0,
+        random_state: Optional[int] = None,
         add_encoders: Optional[dict] = None,
     ):
         """ARIMA
@@ -81,7 +81,11 @@ def __init__(
         self.seasonal_order = seasonal_order
         self.trend = trend
         self.model = None
-        np.random.seed(random_state)
+        self._random_state = (
+            random_state
+            if random_state is None
+            else np.random.RandomState(random_state)
+        )
 
     def _fit(self, series: TimeSeries, future_covariates: Optional[TimeSeries] = None):
         super()._fit(series, future_covariates)
@@ -144,6 +148,7 @@ def _predict(
                 nsimulations=n,
                 repetitions=num_samples,
                 initial_state=self.model.states.predicted[-1, :],
+                random_state=self._random_state,
                 exog=future_covariates.values(copy=False)
                 if future_covariates
                 else None,

darts/tests/models/forecasting/test_local_forecasting_models.py

@@ -616,7 +616,7 @@ def test_model_repr_call(self):
             ),  # no params changed
             (
                 ARIMA(1, 1, 1),
-                "ARIMA(p=1, d=1, q=1, seasonal_order=(0, 0, 0, 0), trend=None, random_state=0, add_encoders=None)",
+                "ARIMA(p=1, d=1, q=1, seasonal_order=(0, 0, 0, 0), trend=None, random_state=None, add_encoders=None)",
             ),  # default value for a param
         ]
         for model, expected in model_expected_name_pairs:

darts/tests/models/forecasting/test_probabilistic_models.py

@@ -48,7 +48,7 @@
 
 models_cls_kwargs_errs = [
     (ExponentialSmoothing, {}, 0.3),
-    (ARIMA, {"p": 1, "d": 0, "q": 1}, 0.03),
+    (ARIMA, {"p": 1, "d": 0, "q": 1, "random_state": 42}, 0.03),
 ]
 
 models_cls_kwargs_errs += [
@@ -150,7 +150,6 @@ class ProbabilisticTorchModelsTestCase(DartsBaseTestClass):
     def test_fit_predict_determinism(self):
 
         for model_cls, model_kwargs, _ in models_cls_kwargs_errs:
-
             # whether the first predictions of two models initiated with the same random state are the same
             model = model_cls(**model_kwargs)
             model.fit(self.constant_noisy_ts)

requirements/core.txt

@@ -4,7 +4,7 @@ joblib>=0.16.0
 lightgbm>=3.2.0
 matplotlib>=3.3.0
 nfoursid>=1.0.0
-numpy>=1.19.0
+numpy>=1.19.0,<1.24.0
 pandas>=1.0.5
 pmdarima>=1.8.0
 prophet>=1.1.1