Merge pull request #424 from openclimatefix/issue/390-split

update to split on training in one year, test after a date

Author: peterdudfield

nowcasting_dataset/config/model.py

@@ -292,7 +292,7 @@ class Process(BaseModel):
         ),
     )
     split_method: split.SplitMethod = Field(
-        split.SplitMethod.DAY,
+        split.SplitMethod.DAY_RANDOM_TEST_DATE,
         description=(
             "The method used to split the t0 datetimes into train, validation and test sets."
         ),

nowcasting_dataset/dataset/split/split.py

@@ -49,7 +49,7 @@ class SplitName(Enum):
 def split_data(
     datetimes: Union[List[pd.Timestamp], pd.DatetimeIndex],
     method: SplitMethod,
-    train_test_validation_split: Tuple[int] = (3, 1, 1),
+    train_test_validation_split: Tuple[int, int, int] = (3, 1, 1),
     train_test_validation_specific: TrainValidationTestSpecific = (
         default_train_test_validation_specific
     ),

nowcasting_dataset/manager.py

@@ -166,8 +166,15 @@ def create_files_specifying_spatial_and_temporal_locations_of_each_example_if_ne
         t0_datetimes = self.get_t0_datetimes_across_all_data_sources(
             freq=self.config.process.t0_datetime_frequency
         )
+        # TODO: move hard code values to config file #426
         split_t0_datetimes = split.split_data(
-            datetimes=t0_datetimes, method=self.config.process.split_method
+            datetimes=t0_datetimes,
+            method=self.config.process.split_method,
+            train_test_validation_split=(3, 0, 1),
+            train_validation_test_datetime_split=[
+                pd.Timestamp("2020-01-01"),
+                pd.Timestamp("2021-01-01"),
+            ],
         )
         for split_name, datetimes_for_split in split_t0_datetimes._asdict().items():
             n_batches = self._get_n_batches_for_split_name(split_name)