Merge branch 'main' into xgb-unused-params

Nixtla · Dec 6, 2024 · 2007ee5 · 2007ee5
2 parents 9c3be66 + da8d9e2
commit 2007ee5
Show file tree

Hide file tree

Showing 17 changed files with 339 additions and 532 deletions.
diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
@@ -32,10 +32,10 @@ jobs:
           python-version: ${{ matrix.python-version }}
 
       - name: Install the library
-        run: pip install uv && uv pip install --system ".[all]"
+        run: pip install uv && uv pip install --system ".[all]" "numba>=0.60" shap window-ops
 
       - name: Run all tests
-        run: nbdev_test --n_workers 0 --do_print --timing --skip_file_re 'electricity' --flags 'polars'
+        run: nbdev_test --n_workers 0 --do_print --timing --skip_file_re 'electricity' --flags 'polars shap window_ops'
 
   run-local-tests:
     runs-on: ${{ matrix.os }}
@@ -44,7 +44,7 @@ jobs:
       fail-fast: false
       matrix:
         os: [macos-13, macos-14, windows-latest]
-        python-version: ["3.9", "3.10", "3.11", "3.12"]
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
     steps:
       - name: Clone repo
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -76,7 +76,7 @@ jobs:
           python-version: "3.10"
 
       - name: Install forecast notebook dependencies
-        run: pip install uv && uv pip install --system . datasetsforecast lightgbm matplotlib nbdev xgboost
+        run: pip install uv && uv pip install --system . lightgbm matplotlib nbdev pyarrow xgboost
 
       - name: Run forecast notebook
         run: nbdev_test --path nbs/forecast.ipynb

diff --git a/mlforecast/__init__.py b/mlforecast/__init__.py
@@ -1,3 +1,3 @@
-__version__ = "0.15.1"
+__version__ = "1.0.0"
 __all__ = ['MLForecast']
 from mlforecast.forecast import MLForecast
diff --git a/mlforecast/_modidx.py b/mlforecast/_modidx.py
@@ -494,7 +494,6 @@
                                   'mlforecast.utils._ShortSeriesException': ('utils.html#_shortseriesexception', 'mlforecast/utils.py'),
                                   'mlforecast.utils._ShortSeriesException.__init__': ( 'utils.html#_shortseriesexception.__init__',
                                                                                        'mlforecast/utils.py'),
-                                  'mlforecast.utils._ensure_shallow_copy': ('utils.html#_ensure_shallow_copy', 'mlforecast/utils.py'),
                                   'mlforecast.utils.generate_daily_series': ('utils.html#generate_daily_series', 'mlforecast/utils.py'),
                                   'mlforecast.utils.generate_prices_for_series': ( 'utils.html#generate_prices_for_series',
                                                                                    'mlforecast/utils.py')}}}
diff --git a/mlforecast/compat.py b/mlforecast/compat.py
@@ -22,18 +22,26 @@ def __init__(self, *args, **kwargs):  # noqa: ARG002
             raise ImportError("Please install lightgbm to use this model.")
 
 
-try:
-    from window_ops.shift import shift_array
-except ImportError:
-
-    def shift_array(*_args, **_kwargs):  # noqa: ARG002
-        raise Exception
-
-
 try:
     from xgboost import XGBRegressor
 except ImportError:
 
     class XGBRegressor:
         def __init__(self, *args, **kwargs):  # noqa: ARG002
             raise ImportError("Please install xgboost to use this model.")
+
+
+try:
+    from window_ops.shift import shift_array
+except ImportError:
+    import numpy as np
+    from utilsforecast.compat import njit
+
+    @njit
+    def shift_array(x, offset):
+        if offset >= x.size or offset < 0:
+            return np.full_like(x, np.nan)
+        out = np.empty_like(x)
+        out[:offset] = np.nan
+        out[offset:] = x[:-offset]
+        return out
diff --git a/mlforecast/core.py b/mlforecast/core.py
@@ -46,7 +46,7 @@
     _BaseGroupedArrayTargetTransform,
     BaseTargetTransform,
 )
-from .utils import _ShortSeriesException, _ensure_shallow_copy
+from .utils import _ShortSeriesException
 
 # %% ../nbs/core.ipynb 10
 date_features_dtypes = {
@@ -154,28 +154,19 @@ def _parse_transforms(
         namer = _build_transform_name
     for lag in lags:
         transforms[f"lag{lag}"] = Lag(lag)
-    has_fns = False
     for lag in lag_transforms.keys():
         for tfm in lag_transforms[lag]:
             if isinstance(tfm, _BaseLagTransform):
                 tfm_name = namer(tfm, lag)
                 transforms[tfm_name] = clone(tfm)._set_core_tfm(lag)
             else:
-                has_fns = True
                 tfm, *args = _as_tuple(tfm)
                 assert callable(tfm)
                 tfm_name = namer(tfm, lag, *args)
                 transforms[tfm_name] = (lag, tfm, *args)
-    if has_fns:
-        warnings.warn(
-            "The `window_ops` package (and thus `numba`) will no longer be "
-            "a dependency in a future version.\n"
-            "Please make sure to add it to your requirements to ensure compatibility.",
-            category=FutureWarning,
-        )
     return transforms
 
-# %% ../nbs/core.ipynb 22
+# %% ../nbs/core.ipynb 21
 class TimeSeries:
     """Utility class for storing and transforming time series data."""
 
@@ -512,8 +503,7 @@ def _transform(
             target_names = [f"{self.target_col}{i}" for i in range(max_horizon)]
             df = ufp.assign_columns(df, target_names, target)
         else:
-            if isinstance(df, pd.DataFrame):
-                df = _ensure_shallow_copy(df)
+            df = ufp.copy_if_pandas(df, deep=False)
             df = ufp.assign_columns(df, self.target_col, target)
         return df
 

diff --git a/mlforecast/feature_engineering.py b/mlforecast/feature_engineering.py
@@ -6,27 +6,22 @@
 # %% ../nbs/feature_engineering.ipynb 3
 from typing import Optional
 
-from utilsforecast.compat import DataFrame
-from utilsforecast.processing import (
-    drop_index_if_pandas,
-    horizontal_concat,
-    process_df,
-    take_rows,
-)
+import utilsforecast.processing as ufp
+from utilsforecast.compat import DFType
 from utilsforecast.validation import validate_format
 
 from .core import _parse_transforms, Lags, LagTransforms
 from .grouped_array import GroupedArray
 
 # %% ../nbs/feature_engineering.ipynb 4
 def transform_exog(
-    df: DataFrame,
+    df: DFType,
     lags: Optional[Lags] = None,
     lag_transforms: Optional[LagTransforms] = None,
     id_col: str = "unique_id",
     time_col: str = "ds",
     num_threads: int = 1,
-) -> DataFrame:
+) -> DFType:
     """Compute lag features for dynamic exogenous regressors.
 
     Parameters
@@ -60,11 +55,11 @@ def transform_exog(
     # this is just a dummy target because process_df requires one
     target_col = targets[0]
     validate_format(df, id_col, time_col, target_col)
-    _, _, data, indptr, sort_idxs = process_df(df, id_col, time_col, target_col)
+    processed = ufp.process_df(df, id_col, time_col, target_col)
     results = {}
     cols = []
     for j, target in enumerate(targets):
-        ga = GroupedArray(data[:, j], indptr)
+        ga = GroupedArray(processed.data[:, j], processed.indptr)
         named_tfms = {f"{target}_{k}": v for k, v in tfms.items()}
         if num_threads == 1 or len(named_tfms) == 1:
             computed_tfms = ga.apply_transforms(
@@ -76,9 +71,9 @@ def transform_exog(
             )
         results.update(computed_tfms)
         cols.extend(list(named_tfms.keys()))
-    if sort_idxs is not None:
-        base_df = take_rows(df, sort_idxs)
+    if processed.sort_idxs is not None:
+        base_df = ufp.take_rows(df, processed.sort_idxs)
     else:
         base_df = df
-    base_df = drop_index_if_pandas(base_df)
-    return horizontal_concat([base_df, type(df)(results)[cols]])
+    base_df = ufp.drop_index_if_pandas(base_df)
+    return ufp.horizontal_concat([base_df, type(df)(results)[cols]])
diff --git a/mlforecast/utils.py b/mlforecast/utils.py
@@ -133,15 +133,6 @@ def __repr__(self):
         return f"PredictionIntervals(n_windows={self.n_windows}, h={self.h}, method='{self.method}')"
 
 # %% ../nbs/utils.ipynb 20
-def _ensure_shallow_copy(df: pd.DataFrame) -> pd.DataFrame:
-    from packaging.version import Version
-
-    if Version(pd.__version__) < Version("1.4"):
-        # https://github.com/pandas-dev/pandas/pull/43406
-        df = df.copy()
-    return df
-
-# %% ../nbs/utils.ipynb 21
 class _ShortSeriesException(Exception):
     def __init__(self, idxs):
         self.idxs = idxs
diff --git a/nbs/compat.ipynb b/nbs/compat.ipynb
@@ -37,19 +37,28 @@
     "            )\n",
     "\n",
     "try:\n",
-    "    from window_ops.shift import shift_array\n",
-    "except ImportError:\n",
-    "    def shift_array(*_args, **_kwargs):  # noqa: ARG002\n",
-    "        raise Exception\n",
-    "\n",
-    "try:\n",
     "    from xgboost import XGBRegressor\n",
     "except ImportError:\n",
     "    class XGBRegressor:\n",
     "        def __init__(self, *args, **kwargs):  # noqa: ARG002\n",
     "            raise ImportError(\n",
     "                \"Please install xgboost to use this model.\"\n",
-    "            )"
+    "            )\n",
+    "\n",
+    "try:\n",
+    "    from window_ops.shift import shift_array\n",
+    "except ImportError:\n",
+    "    import numpy as np\n",
+    "    from utilsforecast.compat import njit\n",
+    "\n",
+    "    @njit\n",
+    "    def shift_array(x, offset):\n",
+    "        if offset >= x.size or offset < 0:\n",
+    "            return np.full_like(x, np.nan)\n",
+    "        out = np.empty_like(x)\n",
+    "        out[:offset] = np.nan\n",
+    "        out[offset:] = x[:-offset]\n",
+    "        return out"
    ]
   }
  ],