Added support for calculating common metrics for 2d inputs

ocf_ml_metrics/metrics/errors.py

@@ -25,8 +25,15 @@ def common_metrics(predictions: np.ndarray, target: np.ndarray, tag: str = "", *
     """
     error_dict = {}
 
-    error_dict[tag + "mae"] = np.mean(np.abs(predictions - target))
-    error_dict[tag + "rmse"] = np.sqrt(np.mean(np.square(predictions - target)))
+    def _mean(input):
+        # 2+ dimensional input - compute mean but preserve 0th dimension, yields 1-d ndarray
+        if len(predictions.shape) > 1:
+            return np.mean(input, axis=0)
+        else:
+            return np.mean(input)
+
+    error_dict[tag + "mae"] = _mean(np.abs(predictions - target))
+    error_dict[tag + "rmse"] = np.sqrt(_mean(np.square(predictions - target)))
 
     return error_dict
 

tests/metrics/test_errors.py

@@ -10,15 +10,26 @@
 from tests.consts_for_tests import N_METRICS
 
 
-def test_common_error_metrics():
-    predictions = np.random.random((288, 1))
-    target = np.random.random((288, 1))
+def test_common_error_metrics_1d_input():
+    predictions = np.random.random(8)
+    target = np.random.random(8)
     errors = common_metrics(predictions=predictions, target=target)
     for key in ["mae", "rmse"]:
         assert key in errors.keys()
         assert isinstance(errors[key], float)
 
 
+def test_common_error_metrics_2d_input():
+    predictions = np.random.random((8, 16))
+    target = np.random.random((8, 16))
+    errors = common_metrics(predictions=predictions, target=target)
+    for key in ["mae", "rmse"]:
+        assert key in errors.keys()
+        assert isinstance(errors[key], np.ndarray)
+        assert errors[key].shape == (16,)
+        assert errors[key].dtype == np.float64
+
+
 def test_compute_error_part_of_year():
     predictions = np.zeros((12, 1))