ENH: add mean_squared_log_error (#725)

* ENH: add mean_squared_log_error

Author: raybellwaves

dask_ml/metrics/__init__.py

@@ -4,5 +4,10 @@
     pairwise_distances,
     pairwise_distances_argmin_min,
 )
-from .regression import mean_absolute_error, mean_squared_error, r2_score  # noqa
+from .regression import (  # noqa
+    mean_absolute_error,
+    mean_squared_error,
+    mean_squared_log_error,
+    r2_score,
+)
 from .scorer import SCORERS, check_scoring, get_scorer  # noqa

dask_ml/metrics/regression.py

@@ -110,3 +110,22 @@ def r2_score(
     if compute:
         result = result.compute()
     return result
+
+
+@derived_from(sklearn.metrics)
+def mean_squared_log_error(
+    y_true: ArrayLike,
+    y_pred: ArrayLike,
+    sample_weight: Optional[ArrayLike] = None,
+    multioutput: Optional[str] = "uniform_average",
+    compute: bool = True,
+) -> ArrayLike:
+
+    result = mean_squared_error(
+        np.log1p(y_true),
+        np.log1p(y_pred),
+        sample_weight=sample_weight,
+        multioutput=multioutput,
+        compute=compute,
+    )
+    return result

docs/source/modules/api.rst

@@ -231,6 +231,7 @@ Regression Metrics
 
    metrics.mean_absolute_error
    metrics.mean_squared_error
+   metrics.mean_squared_log_error
    metrics.r2_score
 
 

tests/metrics/test_regression.py

@@ -48,3 +48,15 @@ def test_mse_squared(squared):
     result = m1(a, b, squared=squared)
     expected = m2(a, b, squared=squared)
     assert abs(result - expected) < 1e-5
+
+
+def test_mean_squared_log_error():
+    m1 = dask_ml.metrics.mean_squared_log_error
+    m2 = sklearn.metrics.mean_squared_log_error
+
+    a = da.random.uniform(size=(100,), chunks=(25,))
+    b = da.random.uniform(size=(100,), chunks=(25,))
+
+    result = m1(a, b)
+    expected = m2(a, b)
+    assert abs(result - expected) < 1e-5