Wape metric

CHANGELOG.md

@@ -15,6 +15,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - 
 - 
 - 
+- Add `WAPE` metric & `wape` functional metric ([#1085](https://github.com/tinkoff-ai/etna/pull/1085))
+- 
+- 
 - 
 - 
 ### Changed

etna/metrics/__init__.py

@@ -11,6 +11,7 @@
 from etna.metrics.functional_metrics import rmse
 from etna.metrics.functional_metrics import sign
 from etna.metrics.functional_metrics import smape
+from etna.metrics.functional_metrics import wape
 from etna.metrics.intervals_metrics import Coverage
 from etna.metrics.intervals_metrics import Width
 from etna.metrics.metrics import MAE
@@ -20,6 +21,7 @@
 from etna.metrics.metrics import R2
 from etna.metrics.metrics import RMSE
 from etna.metrics.metrics import SMAPE
+from etna.metrics.metrics import WAPE
 from etna.metrics.metrics import MaxDeviation
 from etna.metrics.metrics import MedAE
 from etna.metrics.metrics import Sign

etna/metrics/functional_metrics.py

@@ -147,3 +147,34 @@ def max_deviation(y_true: ArrayLike, y_pred: ArrayLike) -> float:
 
 
 rmse = partial(mse, squared=False)
+
+
+def wape(y_true: ArrayLike, y_pred: ArrayLike) -> float:
+    """Weighted average percentage Error metric.
+
+    .. math::
+        WAPE(y\_true, y\_pred) = \\frac{\\sum_{i=0}^{n} |y\_true_i - y\_pred_i|}{\\sum_{i=0}^{n}|y\\_true_i|}
+
+    Parameters
+    ----------
+    y_true:
+        array-like of shape (n_samples,) or (n_samples, n_outputs)
+
+        Ground truth (correct) target values.
+
+    y_pred:
+        array-like of shape (n_samples,) or (n_samples, n_outputs)
+
+        Estimated target values.
+
+    Returns
+    -------
+    float
+        A floating point value (the best value is 0.0).
+    """
+    y_true_array, y_pred_array = np.asarray(y_true), np.asarray(y_pred)
+
+    if len(y_true_array.shape) != len(y_pred_array.shape):
+        raise ValueError("Shapes of the labels must be the same")
+
+    return np.sum(np.abs(y_true_array - y_pred_array)) / np.sum(np.abs(y_true_array))

etna/metrics/metrics.py

@@ -8,6 +8,7 @@
 from etna.metrics import rmse
 from etna.metrics import sign
 from etna.metrics import smape
+from etna.metrics import wape
 from etna.metrics.base import Metric
 from etna.metrics.base import MetricAggregationMode
 
@@ -302,4 +303,32 @@ def greater_is_better(self) -> bool:
         return False
 
 
-__all__ = ["MAE", "MSE", "RMSE", "R2", "MSLE", "MAPE", "SMAPE", "MedAE", "Sign", "MaxDeviation"]
+class WAPE(Metric):
+    """Weighted average percentage Error metric with multi-segment computation support.
+
+    .. math::
+        WAPE(y\_true, y\_pred) = \\frac{\\sum_{i=0}^{n} |y\_true_i - y\_pred_i|}{\\sum_{i=0}^{n}|y\\_true_i|}
+    Notes
+    -----
+    You can read more about logic of multi-segment metrics in Metric docs.
+    """
+
+    def __init__(self, mode: str = MetricAggregationMode.per_segment, **kwargs):
+        """Init metric.
+
+        Parameters
+        ----------
+        mode: 'macro' or 'per-segment'
+            metrics aggregation mode
+        kwargs:
+            metric's computation arguments
+        """
+        super().__init__(mode=mode, metric_fn=wape, **kwargs)
+
+    @property
+    def greater_is_better(self) -> bool:
+        """Whether higher metric value is better."""
+        return False
+
+
+__all__ = ["MAE", "MSE", "RMSE", "R2", "MSLE", "MAPE", "SMAPE", "MedAE", "Sign", "MaxDeviation", "WAPE"]

tests/test_metrics/test_functional_metrics.py

@@ -10,6 +10,7 @@
 from etna.metrics import rmse
 from etna.metrics import sign
 from etna.metrics import smape
+from etna.metrics import wape
 
 
 @pytest.fixture()
@@ -39,6 +40,7 @@ def y_pred_1d():
         (r2_score, 0),
         (sign, -1),
         (max_deviation, 2),
+        (wape, 1),
     ),
 )
 def test_all_1d_metrics(metric, right_metrics_value, y_true_1d, y_pred_1d):
@@ -74,6 +76,7 @@ def y_pred_2d():
         (r2_score, 0.0),
         (sign, -1),
         (max_deviation, 4),
+        (wape, 1),
     ),
 )
 def test_all_2d_metrics(metric, right_metrics_value, y_true_2d, y_pred_2d):

tests/test_metrics/test_metrics.py

@@ -12,6 +12,7 @@
 from etna.metrics import rmse
 from etna.metrics import sign
 from etna.metrics import smape
+from etna.metrics import wape
 from etna.metrics.base import MetricAggregationMode
 from etna.metrics.metrics import MAE
 from etna.metrics.metrics import MAPE
@@ -20,6 +21,7 @@
 from etna.metrics.metrics import R2
 from etna.metrics.metrics import RMSE
 from etna.metrics.metrics import SMAPE
+from etna.metrics.metrics import WAPE
 from etna.metrics.metrics import MaxDeviation
 from etna.metrics.metrics import MedAE
 from etna.metrics.metrics import Sign
@@ -41,6 +43,7 @@
         (Sign, "Sign", {}, ""),
         (MaxDeviation, "MaxDeviation", {}, ""),
         (DummyMetric, "DummyMetric", {"alpha": 1.0}, "alpha = 1.0, "),
+        (WAPE, "WAPE", {}, ""),
     ),
 )
 def test_repr(metric_class, metric_class_repr, metric_params, param_repr):
@@ -56,7 +59,7 @@ def test_repr(metric_class, metric_class_repr, metric_params, param_repr):
 
 @pytest.mark.parametrize(
     "metric_class",
-    (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation),
+    (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, WAPE),
 )
 def test_name_class_name(metric_class):
     """Check metrics name property without changing its during inheritance"""
@@ -80,7 +83,7 @@ def test_name_repr(metric_class):
     assert metric_name == true_name
 
 
-@pytest.mark.parametrize("metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation))
+@pytest.mark.parametrize("metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, WAPE))
 def test_metrics_macro(metric_class, train_test_dfs):
     """Check metrics interface in 'macro' mode"""
     forecast_df, true_df = train_test_dfs
@@ -90,7 +93,7 @@ def test_metrics_macro(metric_class, train_test_dfs):
 
 
 @pytest.mark.parametrize(
-    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric)
+    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric, WAPE)
 )
 def test_metrics_per_segment(metric_class, train_test_dfs):
     """Check metrics interface in 'per-segment' mode"""
@@ -103,7 +106,7 @@ def test_metrics_per_segment(metric_class, train_test_dfs):
 
 
 @pytest.mark.parametrize(
-    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric)
+    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric, WAPE)
 )
 def test_metrics_invalid_aggregation(metric_class):
     """Check metrics behavior in case of invalid aggregation mode"""
@@ -112,7 +115,7 @@ def test_metrics_invalid_aggregation(metric_class):
 
 
 @pytest.mark.parametrize(
-    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric)
+    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric, WAPE)
 )
 def test_invalid_timestamps(metric_class, two_dfs_with_different_timestamps):
     """Check metrics behavior in case of invalid timeranges"""
@@ -123,7 +126,7 @@ def test_invalid_timestamps(metric_class, two_dfs_with_different_timestamps):
 
 
 @pytest.mark.parametrize(
-    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric)
+    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric, WAPE)
 )
 def test_invalid_segments(metric_class, two_dfs_with_different_segments_sets):
     """Check metrics behavior in case of invalid segments sets"""
@@ -134,7 +137,7 @@ def test_invalid_segments(metric_class, two_dfs_with_different_segments_sets):
 
 
 @pytest.mark.parametrize(
-    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric)
+    "metric_class", (MAE, MSE, RMSE, MedAE, MSLE, MAPE, SMAPE, R2, Sign, MaxDeviation, DummyMetric, WAPE)
 )
 def test_invalid_segments_target(metric_class, train_test_dfs):
     """Check metrics behavior in case of no target column in segment"""
@@ -159,6 +162,7 @@ def test_invalid_segments_target(metric_class, train_test_dfs):
         (Sign, sign),
         (MaxDeviation, max_deviation),
         (DummyMetric, create_dummy_functional_metric()),
+        (WAPE, wape),
     ),
 )
 def test_metrics_values(metric_class, metric_fn, train_test_dfs):
@@ -191,6 +195,7 @@ def test_metrics_values(metric_class, metric_fn, train_test_dfs):
         (Sign(), None),
         (MaxDeviation(), False),
         (DummyMetric(), False),
+        (WAPE(), False),
     ),
 )
 def test_metrics_greater_is_better(metric, greater_is_better):