dont interpolate nwp data

nowcasting_dataset/config/gcp.yaml

@@ -7,7 +7,7 @@ input_data:
   gsp:
     forecast_minutes: 60
     gsp_zarr_path: gs://solar-pv-nowcasting-data/PV/GSP/v2/pv_gsp.zarr
-    history_minutes: 30
+    history_minutes: 60
   nwp:
     forecast_minutes: 60
     history_minutes: 30

nowcasting_dataset/config/on_premises.yaml

@@ -23,6 +23,7 @@ input_data:
       - hcc
     nwp_image_size_pixels: 64
     nwp_zarr_path: /mnt/storage_b/data/ocf/solar_pv_nowcasting/nowcasting_dataset_pipeline/NWP/UK_Met_Office/UKV/zarr/UKV__2018-01_to_2019-12__chunks__variable10__init_time1__step1__x548__y704__.zarr
+    history_minutes: 60
 
   #---------------------- PV -------------------
   pv:

nowcasting_dataset/data_sources/data_source.py

@@ -57,7 +57,7 @@ def __post_init__(self):
         assert self.forecast_length >= 0
         assert self.history_minutes % self.sample_period_minutes == 0, (
             f"sample period ({self.sample_period_minutes}) minutes "
-            f"does not fit into historic minutes ({self.forecast_minutes})"
+            f"does not fit into historic minutes ({self.history_minutes})"
         )
         assert self.forecast_minutes % self.sample_period_minutes == 0, (
             f"sample period ({self.sample_period_minutes}) minutes "

nowcasting_dataset/data_sources/nwp/nwp_data_source.py

@@ -114,8 +114,7 @@ def _post_process_example(
         """Resamples to 5 minutely."""
         start_dt = self._get_start_dt(t0_dt)
         end_dt = self._get_end_dt(t0_dt)
-        selected_data = selected_data.resample({"target_time": "5T"})
-        selected_data = selected_data.interpolate()
+
         selected_data = selected_data.sel(target_time=slice(start_dt, end_dt))
         selected_data = selected_data.rename({"target_time": "time", "variable": "channels"})
         selected_data.data = selected_data.data.astype(np.float16)
@@ -133,8 +132,12 @@ def datetime_index(self) -> pd.DatetimeIndex:
         target_times = np.unique(target_times)
         target_times = np.sort(target_times)
         target_times = pd.DatetimeIndex(target_times)
-        resampler = pd.Series(0, index=target_times).resample("5T")
-        return resampler.ffill(limit=11).dropna().index
+        return target_times
+
+    @property
+    def sample_period_minutes(self) -> int:
+        """Override the default sample minutes"""
+        return 60
 
 
 def open_nwp(zarr_path: str, consolidated: bool) -> xr.Dataset:

tests/config/test.yaml

@@ -10,6 +10,7 @@ input_data:
       - t
     nwp_image_size_pixels: 2
     nwp_zarr_path: tests/data/nwp_data/test.zarr
+    history_minutes: 60
   pv:
     pv_filename: tests/data/pv_data/test.nc
     pv_metadata_filename: tests/data/pv_metadata/UK_PV_metadata.csv

tests/data_sources/test_nwp_data_source.py

@@ -15,15 +15,15 @@
 def test_nwp_data_source_init():  # noqa: D103
     _ = NWPDataSource(
         zarr_path=NWP_ZARR_PATH,
-        history_minutes=30,
+        history_minutes=60,
         forecast_minutes=60,
     )
 
 
 def test_nwp_data_source_open():  # noqa: D103
     nwp = NWPDataSource(
         zarr_path=NWP_ZARR_PATH,
-        history_minutes=30,
+        history_minutes=60,
         forecast_minutes=60,
         channels=["t"],
     )
@@ -34,7 +34,7 @@ def test_nwp_data_source_open():  # noqa: D103
 def test_nwp_data_source_batch():  # noqa: D103
     nwp = NWPDataSource(
         zarr_path=NWP_ZARR_PATH,
-        history_minutes=30,
+        history_minutes=60,
         forecast_minutes=60,
         channels=["t"],
     )
@@ -47,13 +47,17 @@ def test_nwp_data_source_batch():  # noqa: D103
 
     batch = nwp.get_batch(t0_datetimes=t0_datetimes, x_locations=x, y_locations=y)
 
-    assert batch.data.shape == (4, 1, 19, 2, 2)
+    # batch size 4
+    # channel 1
+    # time series, 1 int he past, 1 now, 1 in the future
+    # x,y of size 2
+    assert batch.data.shape == (4, 1, 3, 2, 2)
 
 
 def test_nwp_get_contiguous_time_periods():  # noqa: D103
     nwp = NWPDataSource(
         zarr_path=NWP_ZARR_PATH,
-        history_minutes=30,
+        history_minutes=60,
         forecast_minutes=60,
         channels=["t"],
     )
@@ -68,13 +72,13 @@ def test_nwp_get_contiguous_time_periods():  # noqa: D103
 def test_nwp_get_contiguous_t0_time_periods():  # noqa: D103
     nwp = NWPDataSource(
         zarr_path=NWP_ZARR_PATH,
-        history_minutes=30,
+        history_minutes=60,
         forecast_minutes=60,
         channels=["t"],
     )
 
     contiguous_time_periods = nwp.get_contiguous_t0_time_periods()
     correct_time_periods = pd.DataFrame(
-        [{"start_dt": pd.Timestamp("2019-01-01 00:30"), "end_dt": pd.Timestamp("2019-01-02 01:00")}]
+        [{"start_dt": pd.Timestamp("2019-01-01 01:00"), "end_dt": pd.Timestamp("2019-01-02 01:00")}]
     )
     pd.testing.assert_frame_equal(contiguous_time_periods, correct_time_periods)