add more verbose message for nans

Author: peterdudfield

nowcasting_dataset/data_sources/datasource_output.py

@@ -58,9 +58,14 @@ def check_nan_and_inf(self, data: xr.Dataset, variable_name: str = None):
         """Check that all values are non NaNs and not infinite"""
 
         if np.isnan(data).any():
-            message = f"Some {self.__class__.__name__} data values are NaNs"
+            message = f"Some {self.__class__.__name__} data values are NaNs. "
             if variable_name is not None:
                 message += f" ({variable_name})"
+
+            # find out which example has nans in it
+            for i in range(data.shape[0]):
+                if np.isnan(data[i]).any():
+                    message += f" Nans in example {i}."
             logger.error(message)
             raise Exception(message)
 

nowcasting_dataset/data_sources/gsp/gsp_data_source.py

@@ -225,8 +225,8 @@ def get_locations(self, t0_datetimes_utc: pd.DatetimeIndex) -> List[SpaceTimeLoc
         if total_gsp_nan_count == 0:
 
             # get random GSP metadata
-            indexes = list(
-                self.rng.integers(low=0, high=len(self.metadata), size=len(t0_datetimes_utc))
+            indexes = sorted(
+                list(self.rng.integers(low=0, high=len(self.metadata), size=len(t0_datetimes_utc)))
             )
             metadata = self.metadata.iloc[indexes]