fix: Updated join key in transform_new function in scale transformer

.gitignore

@@ -405,3 +405,5 @@ $RECYCLE.BIN/
 *.lnk
 
 # End of https://www.toptal.com/developers/gitignore/api/visualstudiocode,pycharm,macos,windows,python,rust,linux,asdf
+requirements-dev.lock
+requirements.lock

functime/base/forecaster.py

@@ -95,8 +95,8 @@ class Forecaster(Model):
     ----------
     freq : str
         Offset alias supported by Polars.
-    lags : int
-        Number of lagged target variables.
+    lags : Optional[int]
+        Number of lagged target variables. Can also be set to 0, this requires `X` to be provided for features.
     max_horizons: Optional[int]
         Maximum number of horizons to predict directly.
         Only applied if `strategy` equals "direct" or "ensemble".
@@ -114,7 +114,7 @@ class Forecaster(Model):
     def __init__(
         self,
         freq: Union[str, None],
-        lags: int,
+        lags: Optional[int],
         max_horizons: Optional[int] = None,
         strategy: FORECAST_STRATEGIES = None,
         target_transform: Optional[Union[Transformer, List[Transformer]]] = None,
@@ -290,7 +290,7 @@ def backtest(
 
         y = check_backtest_lengths(
             y,
-            max_lags=self.lags,
+            max_lags=self.lags if self.lags else 0,
             test_size=test_size,
             drop_short=drop_short,
             drop_tolerance=drop_tolerance,

functime/forecasting/_ar.py

@@ -23,7 +23,7 @@
 
 def fit_recursive(
     regress: Callable[[pl.LazyFrame, pl.LazyFrame], Any],
-    lags: int,
+    lags: Optional[int],
     y: pl.LazyFrame,
     X: Optional[pl.LazyFrame] = None,
 ) -> Mapping[str, Any]:
@@ -39,7 +39,7 @@ def fit_recursive(
     # 2. Fit
     fitted_regressor = regress(X=X_final, y=y_final)
     # 3. Collect artifacts
-    y_lag = make_y_lag(X_y_final, target_col=y.columns[-1], lags=lags)
+    y_lag = make_y_lag(X_y_final, target_col=y.columns[-1], lags=lags if lags else 0)
     artifacts = {
         "regressor": fitted_regressor,
         "y_lag": y_lag.collect(streaming=True),
@@ -79,7 +79,7 @@ def fit_direct(
 
 def fit_autoreg(
     regress: Callable[[pl.LazyFrame, pl.LazyFrame], Any],
-    lags: int,
+    lags: Optional[int],
     y: Union[pl.DataFrame, pl.LazyFrame],
     X: Optional[Union[pl.DataFrame, pl.LazyFrame]] = None,
     max_horizons: Optional[int] = None,
@@ -229,10 +229,15 @@ def predict_recursive(
     lag_cols = y_lag.columns[2:]
     lead_col = lag_cols[0]
 
+    Y_LAG_BLUEPRINT = y_lag
+
     def _get_x_y_slice(y_lag: pl.DataFrame, i: int):
-        x_y_slice = y_lag.select(
-            [entity_col, pl.all().exclude(entity_col).list.get(-1)]
-        )
+        if lead_col[-1] == "0":  # checks if there are no lags
+            x_y_slice = Y_LAG_BLUEPRINT
+        else:
+            x_y_slice = y_lag.select(
+                [entity_col, pl.all().exclude(entity_col).list.get(-1)]
+            )
         if X is not None:
             x = X.select([entity_col, pl.all().exclude(entity_col).list.get(i)])
             x_y_slice = x_y_slice.join(x, on=entity_col, how="left")

functime/forecasting/_reduction.py

@@ -26,10 +26,13 @@ def make_reduction(
     y = y.lazy()
     X = X.lazy() if X is not None else X
     # Get lags
-    y_lag = y.pipe(lag(lags=list(range(1, lags + 1))))
-    X_y = y_lag.join(y, on=idx_cols, how="inner").select(
-        [*y.columns, *y_lag.columns[2:]]
-    )
+    if lags:
+        y_lag = y.pipe(lag(lags=list(range(1, lags + 1))))
+        X_y = y_lag.join(y, on=idx_cols, how="inner").select(
+            [*y.columns, *y_lag.columns[2:]]
+        )
+    else:
+        X_y = y.with_columns(pl.lit(None).alias(f"{y.columns[-1]}__lag_0"))
     # Exogenous features
     if X is not None:
         X_y = _join_X_y(X_y, X)

functime/preprocessing.py

@@ -401,17 +401,18 @@ def invert(state: ModelState, X: pl.LazyFrame) -> pl.LazyFrame:
         return X
 
     def transform_new(state: ModelState, X: pl.LazyFrame) -> pl.LazyFrame:
-        artifacts = state.artifacts
         idx_cols = X.columns[:2]
-        numeric_cols = state.artifacts["numeric_cols"]
-        _mean = artifacts["_mean"]
-        _std = artifacts["_std"]
+        entity_col = idx_cols[0]
+        artifacts = state.artifacts
+        numeric_cols = artifacts["numeric_cols"]
         if use_mean:
-            X = X.join(_mean, on=idx_cols, how="left").select(
+            _mean = artifacts["_mean"]
+            X = X.join(_mean, on=entity_col, how="left").select(
                 idx_cols + [pl.col(col) - pl.col(f"{col}_mean") for col in numeric_cols]
             )
         if use_std:
-            X = X.join(_std, on=idx_cols, how="left").select(
+            _std = artifacts["_std"]
+            X = X.join(_std, on=entity_col, how="left").select(
                 idx_cols + [pl.col(col) / pl.col(f"{col}_std") for col in numeric_cols]
             )
         if rescale_bool:
@@ -1074,4 +1075,4 @@ def transform(X: pl.LazyFrame) -> pl.LazyFrame:
         )
         return {"X_new": X_new}
 
-    return transform
+    return transform