Add NaiveMovingAverage baseline model

darts/models/__init__.py

@@ -10,7 +10,12 @@
 # Forecasting
 from darts.models.forecasting.arima import ARIMA
 from darts.models.forecasting.auto_arima import AutoARIMA
-from darts.models.forecasting.baselines import NaiveDrift, NaiveMean, NaiveSeasonal
+from darts.models.forecasting.baselines import (
+    NaiveDrift,
+    NaiveMean,
+    NaiveMovingAverage,
+    NaiveSeasonal,
+)
 from darts.models.forecasting.exponential_smoothing import ExponentialSmoothing
 from darts.models.forecasting.fft import FFT
 from darts.models.forecasting.kalman_forecaster import KalmanForecaster
@@ -134,7 +139,7 @@ class NotImportedXGBModel:
 from darts.models.filtering.kalman_filter import KalmanFilter
 
 # Filtering
-from darts.models.filtering.moving_average import MovingAverage
+from darts.models.filtering.moving_average_filter import MovingAverageFilter
 from darts.models.forecasting.baselines import NaiveEnsembleModel
 
 # Ensembling

darts/models/filtering/moving_average.py → darts/models/filtering/moving_average_filter.py

@@ -7,7 +7,7 @@
 from darts.timeseries import TimeSeries
 
 
-class MovingAverage(FilteringModel):
+class MovingAverageFilter(FilteringModel):
     """
     A simple moving average filter. Works on deterministic and stochastic series.
     """

darts/models/forecasting/baselines.py

@@ -123,6 +123,57 @@ def predict(self, n: int, num_samples: int = 1, verbose: bool = False):
         return self._build_forecast_series(forecast)
 
 
+class NaiveMovingAverage(LocalForecastingModel):
+    def __init__(self, input_chunk_length: int = 1):
+        """Naive Moving Average Model
+
+        This model forecasts using an auto-regressive moving average (ARMA).
+
+        Parameters
+        ----------
+        input_chunk_length
+            The size of the sliding window used to calculate the moving average
+        """
+        super().__init__()
+        self.input_chunk_length = input_chunk_length
+        self.rolling_window = None
+
+    @property
+    def min_train_series_length(self):
+        return self.input_chunk_length
+
+    def __str__(self):
+        return f"NaiveMovingAverage({self.input_chunk_length})"
+
+    def fit(self, series: TimeSeries):
+        super().fit(series)
+        raise_if_not(
+            series.is_deterministic,
+            "This model expects deterministic time series",
+            logger,
+        )
+
+        self.rolling_window = series[-self.input_chunk_length :].values(copy=False)
+        return self
+
+    def predict(self, n: int, num_samples: int = 1, verbose: bool = False):
+        super().predict(n, num_samples)
+
+        predictions_with_observations = np.concatenate(
+            (self.rolling_window, np.zeros(shape=(n, self.rolling_window.shape[1]))),
+            axis=0,
+        )
+        rolling_sum = sum(self.rolling_window)
+
+        chunk_length = self.input_chunk_length
+        for i in range(chunk_length, chunk_length + n):
+            prediction = rolling_sum / chunk_length
+            predictions_with_observations[i] = prediction
+            lost_value = predictions_with_observations[i - chunk_length]
+            rolling_sum += prediction - lost_value
+        return self._build_forecast_series(predictions_with_observations[-n:])
+
+
 class NaiveEnsembleModel(EnsembleModel):
     def __init__(
         self, models: Union[List[LocalForecastingModel], List[GlobalForecastingModel]]

darts/tests/ad/test_aggregators.py

@@ -7,7 +7,7 @@
 from darts.ad.aggregators.and_aggregator import AndAggregator
 from darts.ad.aggregators.ensemble_sklearn_aggregator import EnsembleSklearnAggregator
 from darts.ad.aggregators.or_aggregator import OrAggregator
-from darts.models import MovingAverage
+from darts.models import MovingAverageFilter
 from darts.tests.base_test_class import DartsBaseTestClass
 
 list_NonFittableAggregator = [
@@ -725,7 +725,7 @@ def test_EnsembleSklearn(self):
 
         # Need to input an EnsembleSklearn model
         with self.assertRaises(ValueError):
-            EnsembleSklearnAggregator(model=MovingAverage(window=10))
+            EnsembleSklearnAggregator(model=MovingAverageFilter(window=10))
 
         # simple case
         # series has 3 components, and real_anomalies_3w is equal to

darts/tests/ad/test_anomaly_model.py

@@ -29,7 +29,7 @@
     WassersteinScorer,
 )
 from darts.ad.utils import eval_accuracy_from_scores, show_anomalies_from_scores
-from darts.models import MovingAverage, NaiveSeasonal, RegressionModel
+from darts.models import MovingAverageFilter, NaiveSeasonal, RegressionModel
 from darts.tests.base_test_class import DartsBaseTestClass
 
 
@@ -59,8 +59,8 @@ class ADAnomalyModelTestCase(DartsBaseTestClass):
         train._time_index, np_only_0_anomalies
     )
 
-    modified_train = MovingAverage(window=10).filter(train)
-    modified_test = MovingAverage(window=10).filter(test)
+    modified_train = MovingAverageFilter(window=10).filter(train)
+    modified_test = MovingAverageFilter(window=10).filter(test)
 
     np_probabilistic = np.random.normal(loc=10, scale=1, size=[100, 1, 20])
     probabilistic = TimeSeries.from_times_and_values(
@@ -95,7 +95,9 @@ def test_Scorer(self):
         for scorers in list_NonFittableAnomalyScorer:
             for anomaly_model in [
                 ForecastingAnomalyModel(model=RegressionModel(lags=10), scorer=scorers),
-                FilteringAnomalyModel(model=MovingAverage(window=20), scorer=scorers),
+                FilteringAnomalyModel(
+                    model=MovingAverageFilter(window=20), scorer=scorers
+                ),
             ]:
 
                 # scorer are trainable
@@ -110,7 +112,9 @@ def test_Scorer(self):
         for scorers in list_FittableAnomalyScorer:
             for anomaly_model in [
                 ForecastingAnomalyModel(model=RegressionModel(lags=10), scorer=scorers),
-                FilteringAnomalyModel(model=MovingAverage(window=20), scorer=scorers),
+                FilteringAnomalyModel(
+                    model=MovingAverageFilter(window=20), scorer=scorers
+                ),
             ]:
 
                 # scorer are not trainable
@@ -123,7 +127,9 @@ def test_Score(self):
         )
         am1.fit(self.train, allow_model_training=True)
 
-        am2 = FilteringAnomalyModel(model=MovingAverage(window=20), scorer=NormScorer())
+        am2 = FilteringAnomalyModel(
+            model=MovingAverageFilter(window=20), scorer=NormScorer()
+        )
 
         for am in [am1, am2]:
             # Parameter return_model_prediction
@@ -149,12 +155,15 @@ def test_Score(self):
     def test_FitFilteringAnomalyModelInput(self):
 
         for anomaly_model in [
-            FilteringAnomalyModel(model=MovingAverage(window=20), scorer=NormScorer()),
             FilteringAnomalyModel(
-                model=MovingAverage(window=20), scorer=[NormScorer(), KMeansScorer()]
+                model=MovingAverageFilter(window=20), scorer=NormScorer()
+            ),
+            FilteringAnomalyModel(
+                model=MovingAverageFilter(window=20),
+                scorer=[NormScorer(), KMeansScorer()],
             ),
             FilteringAnomalyModel(
-                model=MovingAverage(window=20), scorer=KMeansScorer()
+                model=MovingAverageFilter(window=20), scorer=KMeansScorer()
             ),
         ]:
 
@@ -336,12 +345,15 @@ def test_ScoreForecastingAnomalyModelInput(self):
     def test_ScoreFilteringAnomalyModelInput(self):
 
         for anomaly_model in [
-            FilteringAnomalyModel(model=MovingAverage(window=10), scorer=NormScorer()),
             FilteringAnomalyModel(
-                model=MovingAverage(window=10), scorer=[NormScorer(), KMeansScorer()]
+                model=MovingAverageFilter(window=10), scorer=NormScorer()
+            ),
+            FilteringAnomalyModel(
+                model=MovingAverageFilter(window=10),
+                scorer=[NormScorer(), KMeansScorer()],
             ),
             FilteringAnomalyModel(
-                model=MovingAverage(window=10), scorer=KMeansScorer()
+                model=MovingAverageFilter(window=10), scorer=KMeansScorer()
             ),
         ]:
 
@@ -355,15 +367,18 @@ def test_eval_accuracy(self):
         )
         am1.fit(self.train, allow_model_training=True)
 
-        am2 = FilteringAnomalyModel(model=MovingAverage(window=20), scorer=NormScorer())
+        am2 = FilteringAnomalyModel(
+            model=MovingAverageFilter(window=20), scorer=NormScorer()
+        )
 
         am3 = ForecastingAnomalyModel(
             model=RegressionModel(lags=10), scorer=[NormScorer(), WassersteinScorer()]
         )
         am3.fit(self.train, allow_model_training=True)
 
         am4 = FilteringAnomalyModel(
-            model=MovingAverage(window=20), scorer=[NormScorer(), WassersteinScorer()]
+            model=MovingAverageFilter(window=20),
+            scorer=[NormScorer(), WassersteinScorer()],
         )
         am4.fit(self.train)
 
@@ -500,7 +515,9 @@ def test_ForecastingAnomalyModelInput(self):
         with self.assertRaises(ValueError):
             ForecastingAnomalyModel(model=1, scorer=NormScorer())
         with self.assertRaises(ValueError):
-            ForecastingAnomalyModel(model=MovingAverage(window=10), scorer=NormScorer())
+            ForecastingAnomalyModel(
+                model=MovingAverageFilter(window=10), scorer=NormScorer()
+            )
         with self.assertRaises(ValueError):
             ForecastingAnomalyModel(
                 model=[RegressionModel(lags=10), RegressionModel(lags=5)],
@@ -534,23 +551,24 @@ def test_FilteringAnomalyModelInput(self):
             FilteringAnomalyModel(model=RegressionModel(lags=10), scorer=NormScorer())
         with self.assertRaises(ValueError):
             FilteringAnomalyModel(
-                model=[MovingAverage(window=10), MovingAverage(window=10)],
+                model=[MovingAverageFilter(window=10), MovingAverageFilter(window=10)],
                 scorer=NormScorer(),
             )
 
         # scorer input
         # scorer input must be of type AnomalyScorer
         with self.assertRaises(ValueError):
-            FilteringAnomalyModel(model=MovingAverage(window=10), scorer=1)
+            FilteringAnomalyModel(model=MovingAverageFilter(window=10), scorer=1)
         with self.assertRaises(ValueError):
-            FilteringAnomalyModel(model=MovingAverage(window=10), scorer="str")
+            FilteringAnomalyModel(model=MovingAverageFilter(window=10), scorer="str")
         with self.assertRaises(ValueError):
             FilteringAnomalyModel(
-                model=MovingAverage(window=10), scorer=MovingAverage(window=10)
+                model=MovingAverageFilter(window=10),
+                scorer=MovingAverageFilter(window=10),
             )
         with self.assertRaises(ValueError):
             FilteringAnomalyModel(
-                model=MovingAverage(window=10), scorer=[NormScorer(), "str"]
+                model=MovingAverageFilter(window=10), scorer=[NormScorer(), "str"]
             )
 
     def test_univariate_ForecastingAnomalyModel(self):
@@ -700,7 +718,7 @@ def test_univariate_FilteringAnomalyModel(self):
         )
 
         anomaly_model = FilteringAnomalyModel(
-            model=MovingAverage(window=5),
+            model=MovingAverageFilter(window=5),
             scorer=[
                 NormScorer(),
                 Difference(),
@@ -974,7 +992,7 @@ def test_multivariate__FilteringAnomalyModel(self):
 
         # first case: scorers that return univariate scores
         anomaly_model = FilteringAnomalyModel(
-            model=MovingAverage(window=10),
+            model=MovingAverageFilter(window=10),
             scorer=[
                 NormScorer(component_wise=False),
                 WassersteinScorer(),
@@ -1058,7 +1076,7 @@ def test_multivariate__FilteringAnomalyModel(self):
 
         # second case: scorers that return scorers that have the same width as the input
         anomaly_model = FilteringAnomalyModel(
-            model=MovingAverage(window=10),
+            model=MovingAverageFilter(window=10),
             scorer=[
                 NormScorer(component_wise=True),
                 Difference(),
@@ -1379,7 +1397,7 @@ def test_show_anomalies(self):
         forecasting_anomaly_model.fit(self.train, allow_model_training=True)
 
         filtering_anomaly_model = FilteringAnomalyModel(
-            model=MovingAverage(window=10), scorer=NormScorer()
+            model=MovingAverageFilter(window=10), scorer=NormScorer()
         )
 
         for anomaly_model in [forecasting_anomaly_model, filtering_anomaly_model]:

darts/tests/ad/test_scorers.py

@@ -16,7 +16,7 @@
 )
 from darts.ad.scorers import NormScorer as Norm
 from darts.ad.scorers import PoissonNLLScorer, PyODScorer, WassersteinScorer
-from darts.models import MovingAverage
+from darts.models import MovingAverageFilter
 from darts.tests.base_test_class import DartsBaseTestClass
 
 list_NonFittableAnomalyScorer = [
@@ -70,8 +70,8 @@ class ADAnomalyScorerTestCase(DartsBaseTestClass):
         train._time_index, np_only_0_anomalies
     )
 
-    modified_train = MovingAverage(window=10).filter(train)
-    modified_test = MovingAverage(window=10).filter(test)
+    modified_train = MovingAverageFilter(window=10).filter(train)
+    modified_test = MovingAverageFilter(window=10).filter(test)
 
     np_probabilistic = np.random.normal(loc=10, scale=2, size=[100, 1, 20])
     probabilistic = TimeSeries.from_times_and_values(
@@ -90,8 +90,8 @@ class ADAnomalyScorerTestCase(DartsBaseTestClass):
         mts_train._time_index, np_mts_anomalies
     )
 
-    modified_mts_train = MovingAverage(window=10).filter(mts_train)
-    modified_mts_test = MovingAverage(window=10).filter(mts_test)
+    modified_mts_train = MovingAverageFilter(window=10).filter(mts_train)
+    modified_mts_test = MovingAverageFilter(window=10).filter(mts_test)
 
     np_mts_probabilistic = np.random.normal(
         loc=[[10], [5]], scale=[[1], [1.5]], size=[100, 2, 20]
@@ -1278,7 +1278,7 @@ def test_PyODScorer(self):
 
         # model parameter must be pyod.models typy BaseDetector
         with self.assertRaises(ValueError):
-            PyODScorer(model=MovingAverage(window=10))
+            PyODScorer(model=MovingAverageFilter(window=10))
 
         # component_wise parameter
         # component_wise must be bool

darts/tests/models/filtering/test_filters.py

@@ -7,7 +7,7 @@
 from darts.metrics import rmse
 from darts.models.filtering.gaussian_process_filter import GaussianProcessFilter
 from darts.models.filtering.kalman_filter import KalmanFilter
-from darts.models.filtering.moving_average import MovingAverage
+from darts.models.filtering.moving_average_filter import MovingAverageFilter
 from darts.tests.base_test_class import DartsBaseTestClass
 from darts.utils import timeseries_generation as tg
 
@@ -137,15 +137,15 @@ def test_kalman_given_kf(self):
 
 class MovingAverageTestCase(FilterBaseTestClass):
     def test_moving_average_univariate(self):
-        ma = MovingAverage(window=3, centered=False)
+        ma = MovingAverageFilter(window=3, centered=False)
         sine_ts = tg.sine_timeseries(length=30, value_frequency=0.1)
         sine_filtered = ma.filter(sine_ts)
         self.assertGreater(
             np.mean(np.abs(sine_ts.values())), np.mean(np.abs(sine_filtered.values()))
         )
 
     def test_moving_average_multivariate(self):
-        ma = MovingAverage(window=3)
+        ma = MovingAverageFilter(window=3)
         sine_ts = tg.sine_timeseries(length=30, value_frequency=0.1)
         noise_ts = tg.gaussian_timeseries(length=30) * 0.1
         ts = sine_ts.stack(noise_ts)

darts/tests/models/forecasting/test_local_forecasting_models.py

@@ -26,6 +26,7 @@
     LinearRegressionModel,
     NaiveDrift,
     NaiveMean,
+    NaiveMovingAverage,
     NaiveSeasonal,
     Prophet,
     RandomForest,
@@ -84,6 +85,7 @@
     (NaiveSeasonal(), 32),
     (NaiveMean(), 37),
     (NaiveDrift(), 39),
+    (NaiveMovingAverage(input_chunk_length=5), 34),
 ]
 
 dual_models = [