#1101 Implemented min_train_series_length for Theta and FourTheta (#1111)

* #1101 Implemented min_train_series_length for Theta and FourTheta

* #1103 Placed changelog entry under the Unreleased section

Co-authored-by: Rijk van der Meulen <[email protected]>
Co-authored-by: Julien Herzen <[email protected]>

Author: 3 people

CHANGELOG.md

@@ -4,6 +4,8 @@
 Darts is still in an early development phase and we cannot always guarantee backwards compatibility. Changes that may **break code which uses a previous release of Darts** are marked with a "🔴".
 
 ## [Unreleased](https://github.com/unit8co/darts/tree/master)
+- Implemented the min_train_series_length method for the FourTheta and Theta models that overwrites the minimum default of 3 training samples by 2*seasonal_period when appropriate [#1101](https://github.com/unit8co/darts/pull/1101) by [Rijk van der Meulen](https://github.com/rijkvandermeulen)
+
 [Full Changelog](https://github.com/unit8co/darts/compare/0.20.0...master)
 
 

darts/models/forecasting/theta.py

@@ -168,6 +168,17 @@ def predict(self, n: int, num_samples: int = 1) -> "TimeSeries":
     def __str__(self):
         return f"Theta({self.theta})"
 
+    @property
+    def min_train_series_length(self) -> int:
+        if (
+            self.season_mode != SeasonalityMode.NONE
+            and self.seasonality_period
+            and self.seasonality_period > 1
+        ):
+            return 2 * self.seasonality_period
+        else:
+            return 3
+
 
 class FourTheta(ForecastingModel):
     def __init__(
@@ -457,3 +468,14 @@ def __str__(self):
         return "4Theta(theta:{}, curve:{}, model:{}, seasonality:{})".format(
             self.theta, self.trend_mode, self.model_mode, self.season_mode
         )
+
+    @property
+    def min_train_series_length(self) -> int:
+        if (
+            self.season_mode != SeasonalityMode.NONE
+            and self.seasonality_period
+            and self.seasonality_period > 1
+        ):
+            return 2 * self.seasonality_period
+        else:
+            return 3

darts/tests/models/forecasting/test_4theta.py

@@ -76,3 +76,51 @@ def test_best_model(self):
         self.assertTrue(
             mape(val_series, forecast_best) <= mape(val_series, forecast_random)
         )
+
+    def test_min_train_series_length_with_seasonality(self):
+        seasonality_period = 12
+        fourtheta = FourTheta(
+            model_mode=ModelMode.MULTIPLICATIVE,
+            trend_mode=TrendMode.EXPONENTIAL,
+            season_mode=SeasonalityMode.ADDITIVE,
+            seasonality_period=seasonality_period,
+            normalization=False,
+        )
+        theta = Theta(
+            season_mode=SeasonalityMode.ADDITIVE,
+            seasonality_period=seasonality_period,
+        )
+        self.assertEqual(fourtheta.min_train_series_length, 2 * seasonality_period)
+        self.assertEqual(theta.min_train_series_length, 2 * seasonality_period)
+
+    def test_min_train_series_length_without_seasonality(self):
+        fourtheta = FourTheta(
+            model_mode=ModelMode.MULTIPLICATIVE,
+            trend_mode=TrendMode.EXPONENTIAL,
+            season_mode=SeasonalityMode.ADDITIVE,
+            seasonality_period=None,
+            normalization=False,
+        )
+        theta = Theta(
+            season_mode=SeasonalityMode.ADDITIVE,
+            seasonality_period=None,
+        )
+        self.assertEqual(fourtheta.min_train_series_length, 3)
+        self.assertEqual(theta.min_train_series_length, 3)
+
+    def test_fit_insufficient_train_series_length(self):
+        sine_series = st(length=21, freq="MS")
+        with self.assertRaises(ValueError):
+            fourtheta = FourTheta(
+                model_mode=ModelMode.MULTIPLICATIVE,
+                trend_mode=TrendMode.EXPONENTIAL,
+                season_mode=SeasonalityMode.ADDITIVE,
+                seasonality_period=12,
+            )
+            fourtheta.fit(sine_series)
+        with self.assertRaises(ValueError):
+            theta = Theta(
+                season_mode=SeasonalityMode.ADDITIVE,
+                seasonality_period=12,
+            )
+            theta.fit(sine_series)