Chronos: AutoProphet & AutoARIMA without orca when inference

python/chronos/example/auto_model/autoprophet_nyc_taxi.py

@@ -79,6 +79,10 @@ def get_data(args):
     autoprophet_fit_time = time.time() - start_time
     stop_orca_context()
 
+    # save and load
+    autoprophet.save("autoprophet.ckpt")
+    autoprophet = AutoProphet(load_dir="autoprophet.ckpt")
+
     # evaluate
     auto_searched_mse = autoprophet.evaluate(df_test, metrics=['mse'])[0]
     nonauto_searched_mse = prophet.evaluate(df_test, metrics=['mse'])[0]

python/chronos/src/bigdl/chronos/autots/model/auto_arima.py

@@ -16,12 +16,12 @@
 # limitations under the License.
 #
 
-from bigdl.orca.automl.auto_estimator import AutoEstimator
-from bigdl.chronos.model.arima import ARIMABuilder
-
+import warnings
+from bigdl.chronos.model.arima import ARIMABuilder, ARIMAModel
 
 # -
 
+
 class AutoARIMA:
 
     def __init__(self,
@@ -36,6 +36,7 @@ def __init__(self,
                  cpus_per_trial=1,
                  name="auto_arima",
                  remote_dir=None,
+                 load_dir=None,
                  **arima_config
                  ):
         """
@@ -75,22 +76,29 @@ def __init__(self,
         :param arima_config: Other ARIMA hyperparameters.
 
         """
-        self.search_space = {
-            "p": p,
-            "q": q,
-            "seasonal": seasonal,
-            "P": P,
-            "Q": Q,
-            "m": m,
-        }
-        self.metric = metric
-        model_builder = ARIMABuilder()
-        self.auto_est = AutoEstimator(model_builder=model_builder,
-                                      logs_dir=logs_dir,
-                                      resources_per_trial={
-                                          "cpu": cpus_per_trial},
-                                      remote_dir=remote_dir,
-                                      name=name)
+        if load_dir:
+            self.best_model = ARIMAModel()
+            self.best_model.restore(load_dir)
+        try:
+            from bigdl.orca.automl.auto_estimator import AutoEstimator
+            self.search_space = {
+                "p": p,
+                "q": q,
+                "seasonal": seasonal,
+                "P": P,
+                "Q": Q,
+                "m": m,
+            }
+            self.metric = metric
+            model_builder = ARIMABuilder()
+            self.auto_est = AutoEstimator(model_builder=model_builder,
+                                          logs_dir=logs_dir,
+                                          resources_per_trial={
+                                              "cpu": cpus_per_trial},
+                                          remote_dir=remote_dir,
+                                          name=name)
+        except ImportError:
+            warnings.warn("You need to install `bigdl-orca[automl]` to use `fit` function.")
 
     def fit(self,
             data,

python/chronos/src/bigdl/chronos/autots/model/auto_prophet.py

@@ -17,27 +17,26 @@
 #
 
 import pandas as pd
-
-from bigdl.orca.automl.auto_estimator import AutoEstimator
-import bigdl.orca.automl.hp as hp
-from bigdl.chronos.model.prophet import ProphetBuilder
+import warnings
+from bigdl.chronos.model.prophet import ProphetBuilder, ProphetModel
 
 
 # -
 
 class AutoProphet:
 
     def __init__(self,
-                 changepoint_prior_scale=hp.grid_search([0.005, 0.05, 0.1, 0.5]),
-                 seasonality_prior_scale=hp.grid_search([0.01, 0.1, 1.0, 10.0]),
-                 holidays_prior_scale=hp.loguniform(0.01, 10),
-                 seasonality_mode=hp.choice(['additive', 'multiplicative']),
-                 changepoint_range=hp.uniform(0.8, 0.95),
+                 changepoint_prior_scale=None,
+                 seasonality_prior_scale=None,
+                 holidays_prior_scale=None,
+                 seasonality_mode=None,
+                 changepoint_range=None,
                  metric='mse',
                  logs_dir="/tmp/auto_prophet_logs",
                  cpus_per_trial=1,
                  name="auto_prophet",
                  remote_dir=None,
+                 load_dir=None,
                  **prophet_config
                  ):
         """
@@ -70,24 +69,43 @@ def __init__(self,
         :param remote_dir: String. Remote directory to sync training results and checkpoints. It
             defaults to None and doesn't take effects while running in local. While running in
             cluster, it defaults to "hdfs:///tmp/{name}".
+        :param load_dir: Load the ckpt from load_dir. The value defaults to None.
 
         :param prophet_config: Other Prophet hyperparameters.
         """
-        self.search_space = {
-            "changepoint_prior_scale": changepoint_prior_scale,
-            "seasonality_prior_scale": seasonality_prior_scale,
-            "holidays_prior_scale": holidays_prior_scale,
-            "seasonality_mode": seasonality_mode,
-            "changepoint_range": changepoint_range
-        }
-        self.search_space.update(prophet_config)  # update other configs
-        self.metric = metric
-        model_builder = ProphetBuilder()
-        self.auto_est = AutoEstimator(model_builder=model_builder,
-                                      logs_dir=logs_dir,
-                                      resources_per_trial={"cpu": cpus_per_trial},
-                                      remote_dir=remote_dir,
-                                      name=name)
+        if load_dir:
+            self.best_model = ProphetModel()
+            self.best_model.restore(load_dir)
+        try:
+            from bigdl.orca.automl.auto_estimator import AutoEstimator
+            import bigdl.orca.automl.hp as hp
+            self.search_space = {
+                "changepoint_prior_scale": hp.grid_search([0.005, 0.05, 0.1, 0.5])
+                if changepoint_prior_scale is None
+                else changepoint_prior_scale,
+                "seasonality_prior_scale": hp.grid_search([0.01, 0.1, 1.0, 10.0])
+                if seasonality_prior_scale is None
+                else seasonality_prior_scale,
+                "holidays_prior_scale": hp.loguniform(0.01, 10)
+                if holidays_prior_scale is None
+                else holidays_prior_scale,
+                "seasonality_mode": hp.choice(['additive', 'multiplicative'])
+                if seasonality_mode is None
+                else seasonality_mode,
+                "changepoint_range": hp.uniform(0.8, 0.95)
+                if changepoint_range is None
+                else changepoint_range
+            }
+            self.search_space.update(prophet_config)  # update other configs
+            self.metric = metric
+            model_builder = ProphetBuilder()
+            self.auto_est = AutoEstimator(model_builder=model_builder,
+                                          logs_dir=logs_dir,
+                                          resources_per_trial={"cpu": cpus_per_trial},
+                                          remote_dir=remote_dir,
+                                          name=name)
+        except ImportError:
+            warnings.warn("You need to install `bigdl-orca[automl]` to use `fit` function.")
 
     def fit(self,
             data,

python/chronos/src/bigdl/chronos/forecaster/arima_forecaster.py

@@ -77,7 +77,7 @@ def fit(self, data, validation_data):
         """
         self._check_data(data, validation_data)
         data = data.reshape(-1, 1)
-        validation_data = validation_data.reshape(-1, 1)
+        # validation_data = validation_data.reshape(-1, 1)
         return self.internal.fit_eval(data=data,
                                       validation_data=validation_data,
                                       **self.model_config)

python/chronos/src/bigdl/chronos/model/arima.py

@@ -19,11 +19,10 @@
 from pmdarima.arima import ndiffs
 from pmdarima.arima import nsdiffs
 
-from bigdl.orca.automl.metrics import Evaluator
-from bigdl.orca.automl.model.abstract import BaseModel, ModelBuilder
+from bigdl.chronos.metric.forecast_metrics import Evaluator
 
 
-class ARIMAModel(BaseModel):
+class ARIMAModel:
 
     def __init__(self):
         """
@@ -138,7 +137,7 @@ def evaluate(self, target, x=None, metrics=['mse'], rolling=False):
 
         forecasts = self.predict(horizon=len(target), rolling=rolling)
 
-        return [Evaluator.evaluate(m, target, forecasts) for m in metrics]
+        return Evaluator.evaluate(metrics, target, forecasts, aggregate="mean")
 
     def save(self, checkpoint_file):
         if self.model is None:
@@ -152,7 +151,7 @@ def restore(self, checkpoint_file):
         self.model_init = True
 
 
-class ARIMABuilder(ModelBuilder):
+class ARIMABuilder:
 
     def __init__(self, **arima_config):
         """

python/chronos/src/bigdl/chronos/model/prophet.py

@@ -20,11 +20,10 @@
 from prophet.diagnostics import performance_metrics
 from prophet.diagnostics import cross_validation
 
-from bigdl.orca.automl.metrics import Evaluator
-from bigdl.orca.automl.model.abstract import BaseModel, ModelBuilder
+from bigdl.chronos.metric.forecast_metrics import Evaluator
 
 
-class ProphetModel(BaseModel):
+class ProphetModel:
 
     def __init__(self):
         """
@@ -129,7 +128,8 @@ def evaluate(self, target, data=None, metrics=['mse']):
             raise Exception("Needs to call fit_eval or restore first before calling evaluate")
 
         target_pred = self.model.predict(target)
-        return [Evaluator.evaluate(m, target.y.values, target_pred.yhat.values) for m in metrics]
+        return Evaluator.evaluate(metrics, target.y.values,
+                                  target_pred.yhat.values, aggregate="mean")
 
     def save(self, checkpoint):
         if self.model is None:
@@ -143,7 +143,7 @@ def restore(self, checkpoint):
         self.model_init = True
 
 
-class ProphetBuilder(ModelBuilder):
+class ProphetBuilder:
 
     def __init__(self, **prophet_config):
         """

python/chronos/test/bigdl/chronos/autots/model/test_auto_prophet.py

@@ -48,7 +48,7 @@ def test_auto_prophet_fit(self):
                                    seasonality_prior_scale=hp.loguniform(0.01, 10),
                                    holidays_prior_scale=hp.loguniform(0.01, 10),
                                    seasonality_mode=hp.choice(['additive', 'multiplicative']),
-                                   changepoint_range=hp.uniform(0.8, 0.95)
+                                   changepoint_range=hp.uniform(0.8, 0.95),
                                    )
 
         auto_prophet.fit(data=data,
@@ -101,4 +101,10 @@ def test_auto_prophet_save_load(self):
         with tempfile.TemporaryDirectory() as tmp_dir_name:
             ckpt_name = os.path.join(tmp_dir_name, "json")
             auto_prophet.save(ckpt_name)
+            pred = auto_prophet.predict(horizon=10, freq="D")
             auto_prophet.restore(ckpt_name)
+            pred_old = auto_prophet.predict(horizon=10, freq="D")
+            new_auto_prophet = AutoProphet(load_dir=ckpt_name)
+            pred_new = new_auto_prophet.predict(horizon=10, freq="D")
+            np.testing.assert_almost_equal(pred.yhat.values, pred_new.yhat.values)
+            np.testing.assert_almost_equal(pred.yhat.values, pred_old.yhat.values)

python/orca/src/bigdl/orca/automl/search/ray_tune/ray_tune_search_engine.py

@@ -290,8 +290,9 @@ def train_func(config):
             train_data = ray.get(data_id)
             val_data = ray.get(validation_data_id)
             config = convert_bayes_configs(config).copy()
-            if not isinstance(model_builder, ModelBuilder):
-                raise ValueError(f"You must input a ModelBuilder instance for model_builder")
+            # This check is turned off to support ducking typing
+            # if not isinstance(model_builder, ModelBuilder):
+            #     raise ValueError(f"You must input a ModelBuilder instance for model_builder")
             trial_model = model_builder.build(config)
 
             # no need to call build since it is called the first time fit_eval is called.