chore(deps): bump scipy to >=1.17.1

docs/examples/tabpfn-family.ipynb

@@ -304,7 +304,7 @@
    "source": [
     "## Generate forecast\n",
     "\n",
-    "TabPFN outputs nine fixed quantiles (0.1 through 0.9). Pass `level=[0, 20, 40, 60, 80]` to request all supported prediction intervals.\n"
+    "TabPFN outputs nine fixed quantiles (0.1 through 0.9). Pass `level=[20, 40, 60, 80]` for common decile-style prediction intervals.\n"
    ]
   },
   {
@@ -480,7 +480,7 @@
     }
    ],
    "source": [
-    "level = [0, 20, 40, 60, 80]\n",
+    "level = [20, 40, 60, 80]\n",
     "cv_df = tcf.cross_validation(df=df, h=12, freq=\"MS\", level=level)\n"
    ]
   },

docs/examples/timesfm-family.ipynb

@@ -23,7 +23,7 @@
     "\n",
     "In this example we compare all four generations on event pageview data.\n",
     "\n",
-    "**Prediction intervals:** TimesFM outputs nine fixed quantiles (0.1 through 0.9). Pass `level=[0, 20, 40, 60, 80]` to request all supported prediction intervals; custom levels or quantiles are not supported."
+    "**Prediction intervals:** TimesFM outputs nine fixed quantiles (0.1 through 0.9). Pass `level=[20, 40, 60, 80]` for common decile-style prediction intervals; custom levels or quantiles are not supported."
    ]
   },
   {
@@ -360,7 +360,7 @@
     }
    ],
    "source": [
-    "level = [0, 20, 40, 60, 80]  # required to use all TimesFM-supported quantiles\n",
+    "level = [20, 40, 60, 80]\n",
     "cv_df = tcf.cross_validation(df=df, h=12, level=level)"
    ]
   },

docs/examples/tirex-family.ipynb

@@ -254,7 +254,7 @@
       "metadata": {},
       "outputs": [],
       "source": [
-        "level = [0, 20, 40, 60, 80] # we need to pass 0 level since specific quantiles are required by TiRex\n",
+        "level = [20, 40, 60, 80]\n",
         "cv_df = tcf.cross_validation(df=df, h=12, level=level)\n"
       ]
     },

pyproject.toml

@@ -57,6 +57,7 @@ classifiers = [
 # - transformers>=5 (CVE-2026-1839 fix) requires huggingface-hub>=1; the upper
 # bounds were widened together (transformers<6, huggingface-hub<2). Reaching
 # transformers 5 also requires tfc-t0>=0.2 (older tfc-t0 capped hub<1).
+# - scipy>=1.17.1 on Python>=3.11; scipy<=1.15.3 remains on 3.10 (no 1.17 wheels).
 # - lightning and pytorch-lightning are the same library
 # NOTE: dependencies pinned for faster pip installs
 # versions chosen from uv's dependency resolution
@@ -69,7 +70,7 @@ dependencies = [
   "catboost>=1.2.10",
   "datasets>=4.1.1",
   "fire",
-  "foundationforecast>=0.1.8",
+  "foundationforecast>=0.1.12",
   "fsspec>=2025.9.0",
   "huggingface-hub>=0.36.2,<2.0",
   "hydra-core>=1.3.2",
@@ -93,7 +94,8 @@ dependencies = [
   "prophet>=1.1.7",
   "pydantic-ai>=0.7.0",
   "pytorch-lightning==2.4.0",
-  "scipy<=1.15.3",
+  "scipy<=1.15.3 ; python_full_version < '3.11'",
+  "scipy>=1.17.1 ; python_full_version >= '3.11'",
   "statsforecast>=2.0.2",
   "tensorboard>=2.20.0",
   "torch>=2.10,<2.11 ; python_full_version >= '3.11'",

tests/models/foundation/test_tabpfn.py

@@ -24,7 +24,7 @@
 pytestmark = pytest.mark.models
 
 DEFAULT_QUANTILES = [round(i * 0.1, 1) for i in range(1, 10)]
-DEFAULT_LEVEL = [0, 20, 40, 60, 80]
+DEFAULT_LEVEL = [20, 40, 60, 80]
 
 TABPFN_CASES = [
     pytest.param(TABPFN_V2_MODEL, 4096, "TabPFN-2", id="v2"),

tests/models/test_models.py

@@ -282,7 +282,7 @@ def test_using_quantiles(model):
 
 @pytest.mark.parametrize("model", models)
 def test_using_level(model):
-    level = [0, 20, 40, 60, 80]  # corresponds to qs [0.1, 0.2, ..., 0.9]
+    level = [20, 40, 60, 80]  # decile-style levels (maps to native quantile knots)
     df = generate_series(n_series=2, freq="D")
     _level_unsupported = {
         "AutoLGBM",
@@ -323,7 +323,6 @@ def test_using_level(model):
     assert all(col in fcst_df.columns for col in exp_lv_cols)
     assert not any(("-q-" in col) for col in fcst_df.columns)
     # test monotonicity of levels
-    exp_lv_cols = exp_lv_cols[2:]  # remove level 0
     for c1, c2 in zip(exp_lv_cols[:-1:2], exp_lv_cols[1::2], strict=False):
         if model.alias == "ZeroModel":
             # ZeroModel is a constant model, so all levels should be the same

tests/models/utils/test_forecaster.py

@@ -142,7 +142,7 @@ def test_maybe_convert_level_to_quantiles(n_models, quantiles):
     "n_models,level",
     [
         (1, [80]),
-        (2, [0, 80]),
+        (2, [20, 80]),
         (2, [60, 80]),
     ],
 )
@@ -167,26 +167,19 @@ def test_maybe_convert_quantiles_to_level(n_models, level):
     assert result_df.shape[1] == exp_n_cols
     for model in models:
         for lv in level:
-            if lv == 0:
-                pd.testing.assert_series_equal(
-                    result_df[model],
-                    df[f"{model}-q-50"],
-                    check_names=False,
-                )
-            else:
-                alpha = round(1 - lv / 100, 2)
-                q_lo = int((alpha / 2) * 100)
-                q_hi = int((1 - alpha / 2) * 100)
-                pd.testing.assert_series_equal(
-                    result_df[f"{model}-lo-{lv}"],
-                    df[f"{model}-q-{q_lo}"],
-                    check_names=False,
-                )
-                pd.testing.assert_series_equal(
-                    result_df[f"{model}-hi-{lv}"],
-                    df[f"{model}-q-{q_hi}"],
-                    check_names=False,
-                )
+            alpha = round(1 - lv / 100, 2)
+            q_lo = int((alpha / 2) * 100)
+            q_hi = int((1 - alpha / 2) * 100)
+            pd.testing.assert_series_equal(
+                result_df[f"{model}-lo-{lv}"],
+                df[f"{model}-q-{q_lo}"],
+                check_names=False,
+            )
+            pd.testing.assert_series_equal(
+                result_df[f"{model}-hi-{lv}"],
+                df[f"{model}-q-{q_hi}"],
+                check_names=False,
+            )
     # check that maybe convert level to quantiles returns the same result
     pd.testing.assert_frame_equal(
         df,

tests/test_distributed.py

@@ -178,7 +178,6 @@ def test_cross_validation_spark(simple_models, spark_df):
 
 @pytest.mark.distributed
 def test_using_level_spark(foundation_model, spark_session):
-    # level = [0, 20, 40, 60, 80]  # corresponds to qs [0.1, 0.2, ..., 0.9]
     level: list[int | float] = [20, 80]
     df = generate_series(
         n_series=2, freq="D", max_length=100, static_as_categorical=False
@@ -217,7 +216,6 @@ def test_using_level_spark(foundation_model, spark_session):
     assert all(col in fcst_df_pd.columns for col in exp_lv_cols)
     assert not any(("-q-" in col) for col in fcst_df_pd.columns)
     # test monotonicity of levels
-    exp_lv_cols = exp_lv_cols[2:]  # remove level 0
     for c1, c2 in zip(exp_lv_cols[:-1:2], exp_lv_cols[1::2], strict=False):
         for model in foundation_model:
             if model.alias == "ZeroModel":
@@ -312,7 +310,6 @@ def test_cross_validation_dask(simple_models, dask_df):
 def test_using_level_dask(foundation_model):
     import dask.dataframe as dd
 
-    # level = [0, 20, 40, 60, 80]  # corresponds to qs [0.1, 0.2, ..., 0.9]
     level: list[int | float] = [20, 80]
     n_partitions = 2
     df = generate_series(
@@ -352,7 +349,6 @@ def test_using_level_dask(foundation_model):
     assert all(col in fcst_df_pd.columns for col in exp_lv_cols)
     assert not any(("-q-" in col) for col in fcst_df_pd.columns)
     # test monotonicity of levels
-    exp_lv_cols = exp_lv_cols[2:]  # remove level 0
     for c1, c2 in zip(exp_lv_cols[:-1:2], exp_lv_cols[1::2], strict=False):
         for model in foundation_model:
             if model.alias == "ZeroModel":
@@ -425,7 +421,6 @@ def test_cross_validation_ray(simple_models, ray_dataset):
 def test_using_level_ray(foundation_model):
     import ray
 
-    # level = [0, 20, 40, 60, 80]  # corresponds to qs [0.1, 0.2, ..., 0.9]
     level: list[int | float] = [20, 80]
     df = generate_series(
         n_series=2, freq="D", max_length=100, static_as_categorical=False
@@ -473,7 +468,6 @@ def test_using_level_ray(foundation_model):
     assert all(col in fcst_df_pd.columns for col in exp_lv_cols)
     assert not any(("-q-" in col) for col in fcst_df_pd.columns)
     # test monotonicity of levels
-    exp_lv_cols = exp_lv_cols[2:]  # remove level 0
     for c1, c2 in zip(exp_lv_cols[:-1:2], exp_lv_cols[1::2], strict=False):
         for model in foundation_model:
             if model.alias == "ZeroModel":

timecopilot/forecaster.py

@@ -269,6 +269,8 @@ def _cross_validation_pandas(
         step_size: int | None = None,
         level: list[int | float] | None = None,
         quantiles: list[float] | None = None,
+        *,
+        futr_exog_list: list[str] | None = None,
     ) -> pd.DataFrame:
         """
         Internal pandas-only cross-validation implementation.
@@ -285,6 +287,7 @@ def _cross_validation_pandas(
             step_size=step_size,
             level=level,
             quantiles=quantiles,
+            futr_exog_list=futr_exog_list,
         )
 
     def _cross_validation_distributed(
@@ -296,6 +299,8 @@ def _cross_validation_distributed(
         step_size: int | None = None,
         level: list[int | float] | None = None,
         quantiles: list[float] | None = None,
+        *,
+        futr_exog_list: list[str] | None = None,
         num_partitions: int | None = None,
     ) -> DistributedDataFrame:
         """
@@ -368,6 +373,8 @@ def cross_validation(
         step_size: int | None = None,
         level: list[int | float] | None = None,
         quantiles: list[float] | None = None,
+        *,
+        futr_exog_list: list[str] | None = None,
         num_partitions: int | None = None,
     ) -> AnyDataFrame:
         """
@@ -443,6 +450,7 @@ def cross_validation(
                 step_size=step_size,
                 level=level,
                 quantiles=quantiles,
+                futr_exog_list=futr_exog_list,
                 num_partitions=num_partitions,
             )
 
@@ -455,6 +463,7 @@ def cross_validation(
             step_size=step_size,
             level=level,
             quantiles=quantiles,
+            futr_exog_list=futr_exog_list,
         )
 
     def _detect_anomalies_pandas(

timecopilot/models/utils/forecaster.py

@@ -1,4 +1,5 @@
 import pandas as pd
+from foundationforecast.core.exog.covariates import normalize_exog_strategy
 from foundationforecast.core.forecaster import Forecaster as _Forecaster
 from foundationforecast.core.forecaster import (
     QuantileConverter,
@@ -21,6 +22,36 @@
 class Forecaster(_Forecaster):
     """TimeCopilot Forecaster base (extends foundationforecast)."""
 
+    def _ensure_forecaster_state(self) -> None:
+        if not hasattr(self, "exog_strategy"):
+            self.exog_strategy = normalize_exog_strategy(False)
+        if not hasattr(self, "reuse_loaded_model"):
+            self.reuse_loaded_model = True
+
+    def cross_validation(
+        self,
+        df: pd.DataFrame,
+        h: int,
+        freq: str | None = None,
+        n_windows: int = 1,
+        step_size: int | None = None,
+        level: list[int | float] | None = None,
+        quantiles: list[float] | None = None,
+        *,
+        futr_exog_list: list[str] | None = None,
+    ) -> pd.DataFrame:
+        self._ensure_forecaster_state()
+        return super().cross_validation(
+            df,
+            h,
+            freq=freq,
+            n_windows=n_windows,
+            step_size=step_size,
+            level=level,
+            quantiles=quantiles,
+            futr_exog_list=futr_exog_list,
+        )
+
     @staticmethod
     def _maybe_infer_freq(df: pd.DataFrame, freq: str | None) -> str:
         return maybe_infer_freq(df, freq)