Merge pull request #405 from DHI/peak-ratio-comparercollection

Integration test peak ratio

modelskill/comparison/_utils.py

@@ -50,9 +50,13 @@ def _groupby_df(
     metrics: List[Callable],
     n_min: Optional[int] = None,
 ) -> pd.DataFrame:
-    def calc_metrics(group):
-        row = {}
-        row["n"] = len(group)
+    def calc_metrics(group: pd.DataFrame) -> pd.Series:
+        # set index to time column (in most cases a DatetimeIndex, but not always)
+        group = group.set_index("time")
+
+        # TODO is n a metric or not?
+        row = {"n": len(group)}
+
         for metric in metrics:
             row[metric.__name__] = metric(group.obs_val, group.mod_val)
         return pd.Series(row)

modelskill/metrics.py

@@ -531,7 +531,7 @@ def explained_variance(obs: np.ndarray, model: np.ndarray) -> float:
 
 
 def pr(
-    obs: np.ndarray,
+    obs: pd.Series,
     model: np.ndarray,
     inter_event_level: float = 0.7,
     AAP: int = 2,
@@ -544,7 +544,7 @@ def pr(
 
 def peak_ratio(
     obs: pd.Series,
-    model: pd.Series,
+    model: np.ndarray,
     inter_event_level: float = 0.7,
     AAP: int = 2,
     inter_event_time="36h",
@@ -576,6 +576,7 @@ def peak_ratio(
         return np.nan
     assert isinstance(obs.index, pd.DatetimeIndex)
     time = obs.index
+
     # Calculate number of years
     dt_int = time[1:].values - time[0:-1].values
     dt_int_mode = float(stats.mode(dt_int, keepdims=False)[0]) / 1e9  # in seconds
@@ -1206,4 +1207,5 @@ def _parse_metric(
     return [get_metric(m) for m in metrics]
 
 
-__all__ = list(defined_metrics)
+# TODO add non-metric functions to __all__
+__all__ = [str(m) for m in defined_metrics]

tests/test_comparercollection.py

@@ -22,6 +22,18 @@ def _set_attrs(data: xr.Dataset) -> xr.Dataset:
     return data
 
 
+@pytest.fixture
+def cc_pr() -> modelskill.comparison.Comparer:
+    """Real data to test Peak Ratio from top-to-bottom"""
+    obs = modelskill.PointObservation(
+        "tests/testdata/PR_test_data.dfs0", item="Hs_measured"
+    )
+    mod = modelskill.PointModelResult(
+        "tests/testdata/PR_test_data.dfs0", item="Hs_model"
+    )
+    return modelskill.match(obs, mod)
+
+
 @pytest.fixture
 def pc() -> modelskill.comparison.Comparer:
     """A comparer with fake point data and 2 models"""
@@ -543,3 +555,17 @@ def test_plot_accepts_figsize(cc_plot_function):
     ax = cc_plot_function(figsize=figsize)
     a, b = ax.get_figure().get_size_inches()
     assert a, b == figsize
+
+
+def test_peak_ratio(cc):
+    """Non existent peak ratio"""
+    cc = cc.sel(model="m1")
+    sk = cc.skill(metrics=["peak_ratio"])
+
+    assert sk.loc["fake point obs", "peak_ratio"] == pytest.approx(1.119999999)
+
+
+def test_peak_ratio_2(cc_pr):
+    sk = cc_pr.skill(metrics=["peak_ratio"])
+    assert "peak_ratio" in sk.data.columns
+    assert sk.to_dataframe()["peak_ratio"].values == pytest.approx(1.0799999095653732)

tests/test_metrics.py

@@ -193,7 +193,7 @@ def test_ev():
     assert ev == 1.0
 
 
-def test_pr(obs_series, mod_series):
+def test_pr(obs_series: pd.Series, mod_series: pd.Series) -> None:
     # Obs needs to be a series as the mode of the time index is used.
     # Will use the same data for a real test of ev
     obs = obs_series