Refactor/1080 onehsot regressionmodels

darts/models/forecasting/catboost_model.py

@@ -27,6 +27,7 @@ def __init__(
         likelihood: str = None,
         quantiles: List = None,
         random_state: Optional[int] = None,
+        multi_models: Optional[bool] = True,
         **kwargs,
     ):
         """CatBoost Model
@@ -79,6 +80,9 @@ def __init__(
         random_state
             Control the randomness in the fitting procedure and for sampling.
             Default: ``None``.
+        multi_models
+            If True, a separate model will be trained for each future lag to predict. If False, a single model is
+            trained to predict at step 'output_chunk_length' in the future. Default: True.
         **kwargs
             Additional keyword arguments passed to `catboost.CatBoostRegressor`.
         """
@@ -122,6 +126,7 @@ def __init__(
             lags_future_covariates=lags_future_covariates,
             output_chunk_length=output_chunk_length,
             add_encoders=add_encoders,
+            multi_models=multi_models,
             model=CatBoostRegressor(**kwargs),
         )
 

darts/models/forecasting/gradient_boosted_model.py

@@ -31,6 +31,7 @@ def __init__(
         likelihood: str = None,
         quantiles: List[float] = None,
         random_state: Optional[int] = None,
+        multi_models: Optional[bool] = True,
         **kwargs,
     ):
         """Light Gradient Boosted Model
@@ -81,6 +82,9 @@ def __init__(
         random_state
             Control the randomness in the fitting procedure and for sampling.
             Default: ``None``.
+        multi_models
+            If True, a separate model will be trained for each future lag to predict. If False, a single model is
+            trained to predict at step 'output_chunk_length' in the future. Default: True.
         **kwargs
             Additional keyword arguments passed to `lightgbm.LGBRegressor`.
         """
@@ -109,6 +113,7 @@ def __init__(
             lags_future_covariates=lags_future_covariates,
             output_chunk_length=output_chunk_length,
             add_encoders=add_encoders,
+            multi_models=multi_models,
             model=lgb.LGBMRegressor(**kwargs),
         )
 

darts/models/forecasting/linear_regression_model.py

@@ -29,6 +29,7 @@ def __init__(
         likelihood: str = None,
         quantiles: List[float] = None,
         random_state: Optional[int] = None,
+        multi_models: Optional[bool] = True,
         **kwargs,
     ):
         """Linear regression model.
@@ -83,6 +84,9 @@ def __init__(
             <https://numpy.org/doc/stable/reference/random/generator.html#numpy.random.Generator>`_. Ignored when
             no `likelihood` is set.
             Default: ``None``.
+        multi_models
+            If True, a separate model will be trained for each future lag to predict. If False, a single model is
+            trained to predict at step 'output_chunk_length' in the future. Default: True.
         **kwargs
             Additional keyword arguments passed to `sklearn.linear_model.LinearRegression` (by default), to
             `sklearn.linear_model.PoissonRegressor` (if `likelihood="poisson"`), or to
@@ -117,6 +121,7 @@ def __init__(
             output_chunk_length=output_chunk_length,
             add_encoders=add_encoders,
             model=model,
+            multi_models=multi_models,
         )
 
     def __str__(self):

darts/models/forecasting/random_forest.py

@@ -34,6 +34,7 @@ def __init__(
         add_encoders: Optional[dict] = None,
         n_estimators: Optional[int] = 100,
         max_depth: Optional[int] = None,
+        multi_models: Optional[bool] = True,
         **kwargs,
     ):
         """Random Forest Model
@@ -81,6 +82,9 @@ def __init__(
         max_depth : int
             The maximum depth of the tree. If None, then nodes are expanded until all leaves are pure or until all
             leaves contain less than min_samples_split samples.
+        multi_models
+            If True, a separate model will be trained for each future lag to predict. If False, a single model is
+            trained to predict at step 'output_chunk_length' in the future. Default: True.
         **kwargs
             Additional keyword arguments passed to `sklearn.ensemble.RandomForest`.
         """
@@ -96,6 +100,7 @@ def __init__(
             lags_future_covariates=lags_future_covariates,
             output_chunk_length=output_chunk_length,
             add_encoders=add_encoders,
+            multi_models=multi_models,
             model=RandomForestRegressor(**kwargs),
         )
 

darts/models/forecasting/regression_model.py

@@ -50,6 +50,7 @@ def __init__(
         output_chunk_length: int = 1,
         add_encoders: Optional[dict] = None,
         model=None,
+        multi_models: Optional[bool] = True,
     ):
         """Regression Model
         Can be used to fit any scikit-learn-like regressor class to predict the target time series from lagged values.
@@ -97,6 +98,10 @@ def __init__(
             support multi-output regression for multivariate timeseries, in which case one regressor
             will be used per component in the multivariate series.
             If None, defaults to: ``sklearn.linear_model.LinearRegression(n_jobs=-1)``.
+
+        multi_models
+            If True, a separate model will be trained for each future lag to predict. If False, a single model is
+            trained to predict at step 'output_chunk_length' in the future. Default: True.
         """
 
         super().__init__(add_encoders=add_encoders)
@@ -105,6 +110,7 @@ def __init__(
         self.lags = {}
         self.output_chunk_length = None
         self.input_dim = None
+        self.multi_models = multi_models
 
         # model checks
         if self.model is None:
@@ -221,6 +227,8 @@ def __init__(
         )
         self.output_chunk_length = output_chunk_length
 
+        self.pred_dim = self.output_chunk_length if self.multi_models else 1
+
     @property
     def _model_encoder_settings(self) -> Tuple[int, int, bool, bool]:
         lags_covariates = {
@@ -305,6 +313,7 @@ def _create_lagged_data(
             lags_past_covariates=lags_past_covariates,
             lags_future_covariates=lags_future_covariates,
             max_samples_per_ts=max_samples_per_ts,
+            multi_models=self.multi_models,
         )
 
         return training_samples, training_labels
@@ -398,7 +407,9 @@ def fit(
         }
 
         # if multi-output regression
-        if not series[0].is_univariate or self.output_chunk_length > 1:
+        if not series[0].is_univariate or (
+            self.output_chunk_length > 1 and self.multi_models
+        ):
             # and model isn't wrapped already
             if not isinstance(self.model, MultiOutputRegressor):
                 # check whether model supports multi-output regression natively
@@ -528,6 +539,14 @@ def predict(
             "future": (future_covariates, self.lags.get("future")),
         }
 
+        # prepare one_shot shift and step
+        if self.multi_models:
+            shift = 0
+            step = self.output_chunk_length
+        else:
+            shift = self.output_chunk_length - 1
+            step = 1
+
         # dictionary containing covariate data over time span required for prediction
         covariate_matrices = {}
         # dictionary containing covariate lags relative to minimum covariate lag
@@ -542,13 +561,19 @@ def predict(
                     # calculating first and last prediction time steps
                     first_pred_ts = ts.end_time() + 1 * ts.freq
                     last_pred_ts = (
-                        first_pred_ts
-                        + ((n_pred_steps - 1) * self.output_chunk_length) * ts.freq
+                        (
+                            first_pred_ts
+                            + ((n_pred_steps - 1) * self.output_chunk_length) * ts.freq
+                        )
+                        if self.multi_models
+                        else (first_pred_ts + (n - 1) * ts.freq)
                     )
-                    # calculating first and last required time steps
-                    first_req_ts = first_pred_ts + lags[0] * ts.freq
-                    last_req_ts = last_pred_ts + lags[-1] * ts.freq
 
+                    # calculating first and last required time steps
+                    first_req_ts = (
+                        first_pred_ts + (lags[0] - shift) * ts.freq
+                    )  # shift lags if using one_shot
+                    last_req_ts = last_pred_ts + (lags[-1] - shift) * ts.freq
                     # check for sufficient covariate data
                     raise_if_not(
                         cov.start_time() <= first_req_ts
@@ -578,7 +603,10 @@ def predict(
         series_matrix = None
         if "target" in self.lags:
             series_matrix = np.stack(
-                [ts[self.lags["target"][0] :].values(copy=False) for ts in series]
+                [
+                    ts.values(copy=False)[self.lags["target"][0] - shift :, :]
+                    for ts in series
+                ]
             )
 
         # repeat series_matrix to shape (num_samples * num_series, n_lags, n_components)
@@ -591,7 +619,7 @@ def predict(
         # prediction
         predictions = []
         # t_pred indicates the number of time steps after the first prediction
-        for t_pred in range(0, n, self.output_chunk_length):
+        for t_pred in range(0, n, step):
             np_X = []
             # retrieve target lags
             if "target" in self.lags:
@@ -602,9 +630,9 @@ def predict(
                     else series_matrix
                 )
                 np_X.append(
-                    target_matrix[:, self.lags["target"]].reshape(
-                        len(series) * num_samples, -1
-                    )
+                    target_matrix[
+                        :, [lag - shift for lag in self.lags["target"]]
+                    ].reshape(len(series) * num_samples, -1)
                 )
             # retrieve covariate lags, enforce order (dict only preserves insertion order for python 3.6+)
             for cov_type in ["past", "future"]:
@@ -643,7 +671,8 @@ def _predict_and_sample(
     ) -> np.ndarray:
         prediction = self.model.predict(x, **kwargs)
         k = x.shape[0]
-        return prediction.reshape(k, self.output_chunk_length, -1)
+
+        return prediction.reshape(k, self.pred_dim, -1)
 
     def __str__(self):
         return self.model.__str__()
@@ -694,18 +723,17 @@ def _predict_quantiles(
         X is of shape (n_series * n_samples, n_regression_features)
         """
         k = x.shape[0]
+
         if num_samples == 1:
             # return median
             fitted = self._model_container[0.5]
-            return fitted.predict(x, **kwargs).reshape(k, self.output_chunk_length, -1)
+            return fitted.predict(x, **kwargs).reshape(k, self.pred_dim, -1)
 
         model_outputs = []
         for quantile, fitted in self._model_container.items():
             self.model = fitted
             # model output has shape (n_series * n_samples, output_chunk_length, n_components)
-            model_output = fitted.predict(x, **kwargs).reshape(
-                k, self.output_chunk_length, -1
-            )
+            model_output = fitted.predict(x, **kwargs).reshape(k, self.pred_dim, -1)
             model_outputs.append(model_output)
         model_outputs = np.stack(model_outputs, axis=-1)
         # model_outputs has shape (n_series * n_samples, output_chunk_length, n_components, n_quantiles)
@@ -738,7 +766,7 @@ def _predict_normal(self, x: np.ndarray, num_samples: int, **kwargs) -> np.ndarr
             else:
                 output_slice = model_output[0, :, :]
 
-            return output_slice.reshape(k, self.output_chunk_length, -1)
+            return output_slice.reshape(k, self.pred_dim, -1)
 
         # probabilistic case
         # univariate & single-chunk output
@@ -759,7 +787,7 @@ def _normal_sampling(self, model_output: np.ndarray, n_samples: int) -> np.ndarr
         where the last 2 dimensions are mu and sigma.
         """
         shape = model_output.shape
-        chunk_len = self.output_chunk_length
+        chunk_len = self.pred_dim
 
         # treating each component separately
         mu_sigma_list = [model_output[i, :, :] for i in range(shape[0])]
@@ -783,9 +811,7 @@ def _predict_poisson(self, x: np.ndarray, num_samples: int, **kwargs) -> np.ndar
         """
         k = x.shape[0]
 
-        model_output = self.model.predict(x, **kwargs).reshape(
-            k, self.output_chunk_length, -1
-        )
+        model_output = self.model.predict(x, **kwargs).reshape(k, self.pred_dim, -1)
         if num_samples == 1:
             return model_output