Add support for Neuralforecast (georgia-tech-db#1115)

Adding support for `neuralforecast`. Fixes georgia-tech-db#1112.

```sql
DROP TABLE IF EXISTS AirData;

CREATE TABLE AirData (
    unique_id TEXT(30),
    ds TEXT(30),
    y INTEGER);

LOAD CSV 'data/forecasting/air-passengers.csv' INTO AirData;

DROP FUNCTION IF EXISTS Forecast;

CREATE FUNCTION Forecast FROM
(SELECT unique_id, ds, y FROM AirData)
TYPE Forecasting
PREDICT 'y'
HORIZON 12
LIBRARY 'neuralforecast';

SELECT Forecast(12);
```
One quick issue here is that `neuralforecast` needs `horizon` as a
parameter while training, unlike `statsforecast`. Thus, a better way to
call the UDF would be simply `SELECT Forecast();`, which is currently
unsupported. @xzdandy Please let me know your thoughts.

List of stuff yet to be done:

- [x] Incorporate `neuralforecast`
- [x] Fix `HORIZON` redundancy (UPDATE: Being fixed in georgia-tech-db#1121)
- [x] Reuse model with lower horizon no
- [x] Add support for ~multivariate forecasting~ exogenous variables
- [x] Add tests
- [x] Add docs

---------

Co-authored-by: xzdandy <[email protected]>

docs/source/reference/ai/model-forecasting.rst

@@ -120,11 +120,19 @@ EvaDB's default forecast framework is `statsforecast <https://nixtla.github.io/s
    * - LIBRARY (str, default: 'statsforecast')
      - We can select one of `statsforecast` (default) or `neuralforecast`. `statsforecast` provides access to statistical forecasting methods, while `neuralforecast` gives access to deep-learning based forecasting methods.
    * - MODEL (str, default: 'ARIMA')
+<<<<<<< HEAD
      - If LIBRARY is `statsforecast`, we can select one of ARIMA, ting, ETS, Theta. The default is ARIMA. Check `Automatic Forecasting <https://nixtla.github.io/statsforecast/src/core/models_intro.html#automatic-forecasting>`_ to learn details about these models. If LIBRARY is `neuralforecast`, we can select one of NHITS or NBEATS. The default is NBEATS. Check `NBEATS docs <https://nixtla.github.io/neuralforecast/models.nbeats.html>`_ for details.
    * - AUTO (str, default: 'T')
      - If set to 'T', it enables automatic hyperparameter optimization. Must be set to 'T' for `statsforecast` library. One may set this parameter to `false` if LIBRARY is `neuralforecast` for faster (but less reliable) results.
    * - Frequency (str, default: 'auto')
      - A string indicating the frequency of the data. The common used ones are D, W, M, Y, which respectively represents day-, week-, month- and year- end frequency. The default value is M. Check `pandas available frequencies <https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#offset-aliases>`_ for all available frequencies. If it is not provided, the frequency is attempted to be determined automatically.
+=======
+     - If LIBRARY is `statsforecast`, we can select one of ARIMA, CES, ETS, Theta. The default is ARIMA. Check `Automatic Forecasting <https://nixtla.github.io/statsforecast/src/core/models_intro.html#automatic-forecasting>`_ to learn details about these models. If LIBRARY is `neuralforecast`, we can select one of NHITS or NBEATS. The default is NBEATS. Check `NBEATS docs <https://nixtla.github.io/neuralforecast/models.nbeats.html>`_ for details.
+   * - AUTO (str, default: 'T')
+     - If set to 'T', it enables automatic hyperparameter optimization. Must be set to 'T' for `statsforecast` library. One may set this parameter to `false` if LIBRARY is `neuralforecast` for faster (but less reliable) results.
+   * - Frequency (str, default: 'auto')
+     - A string indicating the frequency of the data. The common used ones are D, W, M, Y, which repestively represents day-, week-, month- and year- end frequency. The default value is M. Check `pandas available frequencies <https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#offset-aliases>`_ for all available frequencies. If it is not provided, the frequency is attempted to be determined automatically.
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
 
 Note: If columns other than the ones required as mentioned above are passed while creating the function, they will be treated as exogenous variables if LIBRARY is `neuralforecast`. Otherwise, they would be ignored.
 
@@ -141,6 +149,7 @@ Below is an example query specifying the above parameters:
    ID 'type'
    Frequency 'W';
 <<<<<<< HEAD
+<<<<<<< HEAD
 
 Below is an example query with `neuralforecast` with `trend` column as exogenous and without automatic hyperparameter optimization:
 
@@ -156,6 +165,8 @@ Below is an example query with `neuralforecast` with `trend` column as exogenous
     FREQUENCY 'M';
 =======
 >>>>>>> 53dafecf (feat: sync master staging (#1050))
+=======
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
 =======
    SELECT Forecast(12) FROM AirData;
 
@@ -210,6 +221,8 @@ Below is an example query specifying the above parameters:
    TIME 'saledate'
    ID 'type'
    Frequency 'W';
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
 
 Below is an example query with `neuralforecast` with `trend` column as exogenous and without automatic hyperparameter optimization:
 
@@ -222,10 +235,23 @@ Below is an example query with `neuralforecast` with `trend` column as exogenous
     PREDICT 'y'
     LIBRARY 'neuralforecast'
     AUTO 'f'
+<<<<<<< HEAD
     FREQUENCY 'M';
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
 <<<<<<< HEAD
 =======
 >>>>>>> 53dafecf (feat: sync master staging (#1050))
 =======
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+<<<<<<< HEAD
+<<<<<<< HEAD
+    FREQUENCY 'M';
+=======
+    FREQUENCY 'M';
+>>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
+>>>>>>> eva-master
+=======
+    FREQUENCY 'M';
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))

evadb/executor/create_function_executor.py

@@ -78,9 +78,12 @@
     try_to_import_ludwig,
     try_to_import_neuralforecast,
     try_to_import_sklearn,
+<<<<<<< HEAD
 <<<<<<< HEAD
     try_to_import_statsforecast,
 =======
+=======
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
 =======
     try_to_import_ludwig,
 >>>>>>> 2dacff69 (feat: sync master staging (#1050))
@@ -94,7 +97,14 @@
     try_to_import_sklearn,
     try_to_import_statsforecast,
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
+<<<<<<< HEAD
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+>>>>>>> eva-master
+=======
+    try_to_import_statsforecast,
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
     try_to_import_torch,
     try_to_import_ultralytics,
     try_to_import_xgboost,
@@ -409,6 +419,7 @@ def handle_ultralytics_function(self):
     def handle_forecasting_function(self):
         """Handle forecasting functions"""
         os.environ["CUDA_VISIBLE_DEVICES"] = ""
+<<<<<<< HEAD
 =======
     def handle_forecasting_function(self):
         """Handle forecasting functions"""
@@ -417,7 +428,13 @@ def handle_forecasting_function(self):
         """Handle forecasting functions"""
         os.environ["CUDA_VISIBLE_DEVICES"] = ""
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
+<<<<<<< HEAD
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+>>>>>>> eva-master
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         aggregated_batch_list = []
         child = self.children[0]
         for batch in child.exec():
@@ -627,7 +644,11 @@ def handle_forecasting_function(self):
             if int(x.split("horizon")[1].split(".pkl")[0]) >= horizon
         ]
         if len(existing_model_files) == 0:
+<<<<<<< HEAD
             logger.info("Training, please wait...")
+=======
+            print("Training, please wait...")
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
             if library == "neuralforecast":
                 model.fit(df=data, val_size=horizon)
             else:

evadb/functions/forecast.py

@@ -81,6 +81,9 @@ def setup(
         time_column_rename: str,
         id_column_rename: str,
 <<<<<<< HEAD
+<<<<<<< HEAD
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
         horizon: int,
         library: str,
 =======
@@ -129,6 +132,9 @@ def setup(self, model_name: str, model_path: str):
         self.horizon = int(horizon)
         self.library = library
 <<<<<<< HEAD
+<<<<<<< HEAD
+=======
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
 
 =======
 
@@ -146,12 +152,15 @@ def setup(self, model_name: str, model_path: str):
     )
 >>>>>>> 2dacff69 (feat: sync master staging (#1050))
     def forward(self, data) -> pd.DataFrame:
+<<<<<<< HEAD
 <<<<<<< HEAD
         if self.library == "statsforecast":
             forecast_df = self.model.predict(h=self.horizon)
         else:
             forecast_df = self.model.predict()
 =======
+=======
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         horizon = list(data.iloc[:, -1])[0]
         assert (
             type(horizon) is int
@@ -169,25 +178,44 @@ def forward(self, data) -> pd.DataFrame:
         self.library = library
 
     def forward(self, data) -> pd.DataFrame:
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
         if self.library == "statsforecast":
             forecast_df = self.model.predict(h=self.horizon)
         else:
             forecast_df = self.model.predict()
+<<<<<<< HEAD
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+<<<<<<< HEAD
+<<<<<<< HEAD
+=======
+>>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
+>>>>>>> eva-master
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         forecast_df.reset_index(inplace=True)
         forecast_df = forecast_df.rename(
             columns={
                 "unique_id": self.id_column_rename,
                 "ds": self.time_column_rename,
                 self.model_name: self.predict_column_rename,
             }
+<<<<<<< HEAD
 <<<<<<< HEAD
         )[: self.horizon * forecast_df["unique_id"].nunique()]
 =======
 <<<<<<< HEAD
         )
+<<<<<<< HEAD
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+=======
+        )[: self.horizon * forecast_df["unique_id"].nunique()]
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         return forecast_df
 =======
 <<<<<<< HEAD

setup.py

@@ -143,6 +143,7 @@ def read(path, encoding="utf-8"):
 forecasting_libs = [
     "statsforecast", # MODEL TRAIN AND FINE TUNING
     "neuralforecast" # MODEL TRAIN AND FINE TUNING
+<<<<<<< HEAD
 ]
 
 imagegen_libs = [
@@ -158,6 +159,8 @@ def read(path, encoding="utf-8"):
 forecasting_libs = [
     "statsforecast", # MODEL TRAIN AND FINE TUNING
     "neuralforecast" # MODEL TRAIN AND FINE TUNING
+=======
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
 ]
 =======
 >>>>>>> 8c5b63dc (release: merge staging into master (#1032))

test/integration_tests/long/test_model_forecasting.py

@@ -61,6 +61,15 @@ def setUpClass(cls):
 =======
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+        create_table_query = """
+            CREATE TABLE AirDataPanel (\
+            unique_id TEXT(30),\
+            ds TEXT(30),\
+            y INTEGER,\
+            trend INTEGER,\
+            ylagged INTEGER);"""
+        execute_query_fetch_all(cls.evadb, create_table_query)
+
         create_table_query = """
             CREATE TABLE HomeData (\
             saledate TEXT(30),\
@@ -93,21 +102,38 @@ def setUpClass(cls):
 <<<<<<< HEAD
 <<<<<<< HEAD
 <<<<<<< HEAD
+<<<<<<< HEAD
 =======
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
 =======
 =======
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+=======
+<<<<<<< HEAD
+<<<<<<< HEAD
+=======
+>>>>>>> eva-master
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         path = f"{EvaDB_ROOT_DIR}/data/forecasting/AirPassengersPanel.csv"
         load_query = f"LOAD CSV '{path}' INTO AirDataPanel;"
         execute_query_fetch_all(cls.evadb, load_query)
 
+<<<<<<< HEAD
 <<<<<<< HEAD
 =======
 >>>>>>> 53dafecf (feat: sync master staging (#1050))
 =======
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
+<<<<<<< HEAD
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+>>>>>>> eva-master
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         path = f"{EvaDB_ROOT_DIR}/data/forecasting/home_sales.csv"
         load_query = f"LOAD CSV '{path}' INTO HomeData;"
         execute_query_fetch_all(cls.evadb, load_query)
@@ -212,13 +238,22 @@ def test_forecast(self):
 <<<<<<< HEAD
 <<<<<<< HEAD
 <<<<<<< HEAD
+<<<<<<< HEAD
 <<<<<<< HEAD
             SELECT AirForecast() order by y;
 =======
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
+=======
 =======
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
             SELECT AirForecast(12) order by y;
+<<<<<<< HEAD
 >>>>>>> 53dafecf (feat: sync master staging (#1050))
+=======
+=======
+            SELECT AirForecast() order by y;
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
         """
         result = execute_query_fetch_all(self.evadb, predict_query)
         self.assertEqual(len(result), 12)

test/unit_tests/binder/test_statement_binder.py

@@ -688,6 +688,7 @@ def test_bind_create_function_should_bind_forecast_with_renaming_columns(self):
             self.assertEqual(create_function_statement.inputs, expected_inputs)
             self.assertEqual(create_function_statement.outputs, expected_outputs)
 
+<<<<<<< HEAD
 <<<<<<< HEAD
 =======
 <<<<<<< HEAD
@@ -738,7 +739,13 @@ def test_bind_create_function_should_raise_forecast_with_unexpected_columns(self
 
 =======
 >>>>>>> 40a10ce1 (Bump v0.3.4+ dev)
+<<<<<<< HEAD
 >>>>>>> 6d6a14c8 (Bump v0.3.4+ dev)
+=======
+>>>>>>> eva-master
+=======
+>>>>>>> e8a181c5 (Add support for Neuralforecast (#1115))
+>>>>>>> ca239aea (Add support for Neuralforecast (#1115))
     def test_bind_create_function_should_raise_forecast_missing_required_columns(self):
         with patch.object(StatementBinder, "bind"):
             create_function_statement = MagicMock()