diff --git a/mlforecast/_modidx.py b/mlforecast/_modidx.py
index ac93701f..29d79bc6 100644
--- a/mlforecast/_modidx.py
+++ b/mlforecast/_modidx.py
@@ -262,6 +262,8 @@
                                                                                             'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Combine.update': ( 'lag_transforms.html#combine.update',
                                                                                          'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms.Combine.update_samples': ( 'lag_transforms.html#combine.update_samples',
+                                                                                                 'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.ExpandingMax': ( 'lag_transforms.html#expandingmax',
                                                                                        'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.ExpandingMean': ( 'lag_transforms.html#expandingmean',
@@ -272,12 +274,16 @@
                                                                                             'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.ExpandingQuantile.__init__': ( 'lag_transforms.html#expandingquantile.__init__',
                                                                                                      'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms.ExpandingQuantile.update_samples': ( 'lag_transforms.html#expandingquantile.update_samples',
+                                                                                                           'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.ExpandingStd': ( 'lag_transforms.html#expandingstd',
                                                                                        'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.ExponentiallyWeightedMean': ( 'lag_transforms.html#exponentiallyweightedmean',
                                                                                                     'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.ExponentiallyWeightedMean.__init__': ( 'lag_transforms.html#exponentiallyweightedmean.__init__',
                                                                                                              'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms.ExponentiallyWeightedMean.update_samples': ( 'lag_transforms.html#exponentiallyweightedmean.update_samples',
+                                                                                                                   'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Lag': ('lag_transforms.html#lag', 'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Lag.__eq__': ( 'lag_transforms.html#lag.__eq__',
                                                                                      'mlforecast/lag_transforms.py'),
@@ -287,6 +293,8 @@
                                                                                         'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Lag._set_core_tfm': ( 'lag_transforms.html#lag._set_core_tfm',
                                                                                             'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms.Lag.update_samples': ( 'lag_transforms.html#lag.update_samples',
+                                                                                             'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Offset': ( 'lag_transforms.html#offset',
                                                                                  'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Offset.__init__': ( 'lag_transforms.html#offset.__init__',
@@ -295,6 +303,8 @@
                                                                                            'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.Offset._set_core_tfm': ( 'lag_transforms.html#offset._set_core_tfm',
                                                                                                'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms.Offset.update_samples': ( 'lag_transforms.html#offset.update_samples',
+                                                                                                'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.RollingMax': ( 'lag_transforms.html#rollingmax',
                                                                                      'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms.RollingMean': ( 'lag_transforms.html#rollingmean',
@@ -327,6 +337,8 @@
                                                                                                                 'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._BaseLagTransform._get_name': ( 'lag_transforms.html#_baselagtransform._get_name',
                                                                                                       'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms._BaseLagTransform._lag': ( 'lag_transforms.html#_baselagtransform._lag',
+                                                                                                 'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._BaseLagTransform._set_core_tfm': ( 'lag_transforms.html#_baselagtransform._set_core_tfm',
                                                                                                           'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._BaseLagTransform.stack': ( 'lag_transforms.html#_baselagtransform.stack',
@@ -337,18 +349,26 @@
                                                                                                       'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._BaseLagTransform.update': ( 'lag_transforms.html#_baselagtransform.update',
                                                                                                    'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms._BaseLagTransform.update_samples': ( 'lag_transforms.html#_baselagtransform.update_samples',
+                                                                                                           'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._ExpandingBase': ( 'lag_transforms.html#_expandingbase',
                                                                                          'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._ExpandingBase.__init__': ( 'lag_transforms.html#_expandingbase.__init__',
                                                                                                   'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms._ExpandingBase.update_samples': ( 'lag_transforms.html#_expandingbase.update_samples',
+                                                                                                        'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._RollingBase': ( 'lag_transforms.html#_rollingbase',
                                                                                        'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._RollingBase.__init__': ( 'lag_transforms.html#_rollingbase.__init__',
                                                                                                 'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms._RollingBase.update_samples': ( 'lag_transforms.html#_rollingbase.update_samples',
+                                                                                                      'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._Seasonal_RollingBase': ( 'lag_transforms.html#_seasonal_rollingbase',
                                                                                                 'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._Seasonal_RollingBase.__init__': ( 'lag_transforms.html#_seasonal_rollingbase.__init__',
                                                                                                          'mlforecast/lag_transforms.py'),
+                                           'mlforecast.lag_transforms._Seasonal_RollingBase.update_samples': ( 'lag_transforms.html#_seasonal_rollingbase.update_samples',
+                                                                                                               'mlforecast/lag_transforms.py'),
                                            'mlforecast.lag_transforms._pascal2camel': ( 'lag_transforms.html#_pascal2camel',
                                                                                         'mlforecast/lag_transforms.py')},
             'mlforecast.lgb_cv': { 'mlforecast.lgb_cv.LightGBMCV': ('lgb_cv.html#lightgbmcv', 'mlforecast/lgb_cv.py'),
diff --git a/mlforecast/core.py b/mlforecast/core.py
index 6726d1bb..cfe9e6b2 100644
--- a/mlforecast/core.py
+++ b/mlforecast/core.py
@@ -432,6 +432,16 @@ def _transform(
             self._dropped_series = None
 
         # once we've computed the features and target we can slice the series
+        update_samples = [
+            getattr(tfm, "update_samples", -1) for tfm in self.transforms.values()
+        ]
+        if (
+            self.keep_last_n is None
+            and update_samples
+            and all(samples > 0 for samples in update_samples)
+        ):
+            # user didn't set keep_last_n and we can infer it from the transforms
+            self.keep_last_n = max(update_samples)
         if self.keep_last_n is not None:
             self.ga = self.ga.take_from_groups(slice(-self.keep_last_n, None))
         del self._restore_idxs, self._sort_idxs
diff --git a/mlforecast/lag_transforms.py b/mlforecast/lag_transforms.py
index e1b7cb8b..63cc7fb1 100644
--- a/mlforecast/lag_transforms.py
+++ b/mlforecast/lag_transforms.py
@@ -68,8 +68,17 @@ def stack(transforms: Sequence["_BaseLagTransform"]) -> "_BaseLagTransform":
         )
         return out
 
+    @property
+    def _lag(self):
+        return self._core_tfm.lag - 1
+
+    @property
+    def update_samples(self) -> int:
+        return -1
+
 # %% ../nbs/lag_transforms.ipynb 6
 class Lag(_BaseLagTransform):
+
     def __init__(self, lag: int):
         self.lag = lag
         self._core_tfm = core_tfms.Lag(lag=lag)
@@ -83,6 +92,10 @@ def _get_name(self, lag: int) -> str:
     def __eq__(self, other):
         return isinstance(other, Lag) and self.lag == other.lag
 
+    @property
+    def update_samples(self) -> int:
+        return self.lag
+
 # %% ../nbs/lag_transforms.ipynb 7
 class _RollingBase(_BaseLagTransform):
     "Rolling statistic"
@@ -100,6 +113,10 @@ def __init__(self, window_size: int, min_samples: Optional[int] = None):
         self.window_size = window_size
         self.min_samples = min_samples
 
+    @property
+    def update_samples(self) -> int:
+        return self._lag + self.window_size
+
 # %% ../nbs/lag_transforms.ipynb 8
 class RollingMean(_RollingBase): ...
 
@@ -149,6 +166,10 @@ def __init__(
         self.window_size = window_size
         self.min_samples = min_samples
 
+    @property
+    def update_samples(self) -> int:
+        return self._lag + self.season_length * self.window_size
+
 # %% ../nbs/lag_transforms.ipynb 11
 class SeasonalRollingMean(_Seasonal_RollingBase): ...
 
@@ -183,6 +204,10 @@ class _ExpandingBase(_BaseLagTransform):
 
     def __init__(self): ...
 
+    @property
+    def update_samples(self) -> int:
+        return 1
+
 # %% ../nbs/lag_transforms.ipynb 14
 class ExpandingMean(_ExpandingBase): ...
 
@@ -200,6 +225,10 @@ class ExpandingQuantile(_ExpandingBase):
     def __init__(self, p: float):
         self.p = p
 
+    @property
+    def update_samples(self) -> int:
+        return -1
+
 # %% ../nbs/lag_transforms.ipynb 16
 class ExponentiallyWeightedMean(_BaseLagTransform):
     """Exponentially weighted average
@@ -212,6 +241,10 @@ class ExponentiallyWeightedMean(_BaseLagTransform):
     def __init__(self, alpha: float):
         self.alpha = alpha
 
+    @property
+    def update_samples(self) -> int:
+        return 1
+
 # %% ../nbs/lag_transforms.ipynb 18
 class Offset(_BaseLagTransform):
     """Shift series before computing transformation
@@ -231,9 +264,14 @@ def _get_name(self, lag: int) -> str:
         return self.tfm._get_name(lag + self.n)
 
     def _set_core_tfm(self, lag: int) -> "Offset":
-        self._core_tfm = clone(self.tfm)._set_core_tfm(lag + self.n)
+        self.tfm = clone(self.tfm)._set_core_tfm(lag + self.n)
+        self._core_tfm = self.tfm._core_tfm
         return self
 
+    @property
+    def update_samples(self) -> int:
+        return self.tfm.update_samples + self.n
+
 # %% ../nbs/lag_transforms.ipynb 20
 class Combine(_BaseLagTransform):
     """Combine two lag transformations using an operator
@@ -269,3 +307,7 @@ def transform(self, ga: CoreGroupedArray) -> np.ndarray:
 
     def update(self, ga: CoreGroupedArray) -> np.ndarray:
         return self.operator(self.tfm1.update(ga), self.tfm2.update(ga))
+
+    @property
+    def update_samples(self):
+        return max(self.tfm1.update_samples, self.tfm2.update_samples)
diff --git a/nbs/core.ipynb b/nbs/core.ipynb
index 1cb2fca6..1fefe533 100644
--- a/nbs/core.ipynb
+++ b/nbs/core.ipynb
@@ -20,7 +20,16 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
+     ]
+    }
+   ],
    "source": [
     "#|hide\n",
     "%load_ext autoreload\n",
@@ -916,8 +925,18 @@
     "            self._dropped_series = None\n",
     "\n",
     "        # once we've computed the features and target we can slice the series\n",
+    "        update_samples = [\n",
+    "            getattr(tfm, 'update_samples', -1) for tfm in self.transforms.values()\n",
+    "        ]\n",
+    "        if (\n",
+    "            self.keep_last_n is None\n",
+    "            and update_samples\n",
+    "            and all(samples > 0 for samples in update_samples)\n",
+    "        ):\n",
+    "            # user didn't set keep_last_n and we can infer it from the transforms\n",
+    "            self.keep_last_n = max(update_samples)\n",
     "        if self.keep_last_n is not None:\n",
-    "            self.ga = self.ga.take_from_groups(slice(-self.keep_last_n, None))         \n",
+    "            self.ga = self.ga.take_from_groups(slice(-self.keep_last_n, None))\n",
     "        del self._restore_idxs, self._sort_idxs\n",
     "\n",
     "        # lag transforms\n",
@@ -1671,7 +1690,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L486){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L487){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "## TimeSeries.fit_transform\n",
        "\n",
@@ -1692,7 +1711,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L486){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L487){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "## TimeSeries.fit_transform\n",
        "\n",
@@ -1978,7 +1997,43 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/markdown": [
+       "---\n",
+       "\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L732){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "\n",
+       "## TimeSeries.predict\n",
+       "\n",
+       ">      TimeSeries.predict (models:Dict[str,Union[sklearn.base.BaseEstimator,List\n",
+       ">                          [sklearn.base.BaseEstimator]]], horizon:int,\n",
+       ">                          before_predict_callback:Optional[Callable]=None,\n",
+       ">                          after_predict_callback:Optional[Callable]=None,\n",
+       ">                          X_df:Optional[~DFType]=None,\n",
+       ">                          ids:Optional[List[str]]=None)"
+      ],
+      "text/plain": [
+       "---\n",
+       "\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L732){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "\n",
+       "## TimeSeries.predict\n",
+       "\n",
+       ">      TimeSeries.predict (models:Dict[str,Union[sklearn.base.BaseEstimator,List\n",
+       ">                          [sklearn.base.BaseEstimator]]], horizon:int,\n",
+       ">                          before_predict_callback:Optional[Callable]=None,\n",
+       ">                          after_predict_callback:Optional[Callable]=None,\n",
+       ">                          X_df:Optional[~DFType]=None,\n",
+       ">                          ids:Optional[List[str]]=None)"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "show_doc(TimeSeries.predict, title_level=2)"
    ]
@@ -2094,7 +2149,41 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/markdown": [
+       "---\n",
+       "\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L837){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "\n",
+       "## TimeSeries.update\n",
+       "\n",
+       ">      TimeSeries.update\n",
+       ">                         (df:Union[pandas.core.frame.DataFrame,polars.dataframe\n",
+       ">                         .frame.DataFrame])\n",
+       "\n",
+       "*Update the values of the stored series.*"
+      ],
+      "text/plain": [
+       "---\n",
+       "\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/core.py#L837){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "\n",
+       "## TimeSeries.update\n",
+       "\n",
+       ">      TimeSeries.update\n",
+       ">                         (df:Union[pandas.core.frame.DataFrame,polars.dataframe\n",
+       ">                         .frame.DataFrame])\n",
+       "\n",
+       "*Update the values of the stored series.*"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "show_doc(TimeSeries.update, title_level=2)"
    ]
@@ -2180,7 +2269,15 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "sys:1: CategoricalRemappingWarning: Local categoricals have different encodings, expensive re-encoding is done to perform this merge operation. Consider using a StringCache or an Enum type if the categories are known in advance\n"
+     ]
+    }
+   ],
    "source": [
     "#| hide\n",
     "#| polars\n",
@@ -2429,6 +2526,42 @@
     "preds2 = ts2.predict({'model': NaiveModel()}, 10)\n",
     "pd.testing.assert_frame_equal(preds, preds2)"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| hide\n",
+    "# automatically set keep_last_n for built-in lag transforms\n",
+    "ts = TimeSeries(\n",
+    "    freq='D',\n",
+    "    lags=[1, 2],\n",
+    "    date_features=['dayofweek'],\n",
+    "    lag_transforms={\n",
+    "        1: [RollingMean(1), RollingMean(4)],\n",
+    "    },\n",
+    ")\n",
+    "ts.fit_transform(series, 'unique_id', 'ds', 'y', keep_last_n=20)\n",
+    "assert ts.keep_last_n == 20\n",
+    "ts.fit_transform(series, 'unique_id', 'ds', 'y')\n",
+    "assert ts.keep_last_n == 4\n",
+    "# we can't infer it for functions\n",
+    "ts = TimeSeries(\n",
+    "    freq='D',\n",
+    "    lags=[1, 2],\n",
+    "    date_features=['dayofweek'],\n",
+    "    lag_transforms={\n",
+    "        1: [RollingMean(1), RollingMean(4)],\n",
+    "        5: [expanding_mean],\n",
+    "    },\n",
+    ")\n",
+    "ts.fit_transform(series, 'unique_id', 'ds', 'y', keep_last_n=20)\n",
+    "assert ts.keep_last_n == 20\n",
+    "ts.fit_transform(series, 'unique_id', 'ds', 'y')\n",
+    "assert ts.keep_last_n is None"
+   ]
   }
  ],
  "metadata": {
diff --git a/nbs/forecast.ipynb b/nbs/forecast.ipynb
index d9fc0699..56afd42e 100644
--- a/nbs/forecast.ipynb
+++ b/nbs/forecast.ipynb
@@ -1306,7 +1306,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L127){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L126){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast\n",
        "\n",
@@ -1337,7 +1337,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L127){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L126){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast\n",
        "\n",
@@ -1433,7 +1433,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L447){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L446){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.fit\n",
        "\n",
@@ -1467,7 +1467,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L447){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L446){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.fit\n",
        "\n",
@@ -1548,7 +1548,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L966){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L965){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.save\n",
        "\n",
@@ -1564,7 +1564,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L966){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L965){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.save\n",
        "\n",
@@ -1598,7 +1598,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L983){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L982){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.load\n",
        "\n",
@@ -1614,7 +1614,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L983){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L982){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.load\n",
        "\n",
@@ -1648,7 +1648,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L1006){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L1005){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.update\n",
        "\n",
@@ -1666,7 +1666,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L1006){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L1005){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.update\n",
        "\n",
@@ -1702,7 +1702,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L576){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L575){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.make_future_dataframe\n",
        "\n",
@@ -1718,7 +1718,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L576){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L575){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.make_future_dataframe\n",
        "\n",
@@ -1826,40 +1826,36 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L600){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L599){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.get_missing_future\n",
        "\n",
-       ">      MLForecast.get_missing_future (h:int,\n",
-       ">                                     X_df:Union[pandas.core.frame.DataFrame,pol\n",
-       ">                                     ars.dataframe.frame.DataFrame])\n",
+       ">      MLForecast.get_missing_future (h:int, X_df:~DFType)\n",
        "\n",
        "*Get the missing id and time combinations in `X_df`.*\n",
        "\n",
        "|    | **Type** | **Details** |\n",
        "| -- | -------- | ----------- |\n",
        "| h | int | Number of periods to predict. |\n",
-       "| X_df | Union | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
-       "| **Returns** | **Union** | **DataFrame with expected ids and future times missing in `X_df`** |"
+       "| X_df | DFType | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
+       "| **Returns** | **DFType** | **DataFrame with expected ids and future times missing in `X_df`** |"
       ],
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L600){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L599){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.get_missing_future\n",
        "\n",
-       ">      MLForecast.get_missing_future (h:int,\n",
-       ">                                     X_df:Union[pandas.core.frame.DataFrame,pol\n",
-       ">                                     ars.dataframe.frame.DataFrame])\n",
+       ">      MLForecast.get_missing_future (h:int, X_df:~DFType)\n",
        "\n",
        "*Get the missing id and time combinations in `X_df`.*\n",
        "\n",
        "|    | **Type** | **Details** |\n",
        "| -- | -------- | ----------- |\n",
        "| h | int | Number of periods to predict. |\n",
-       "| X_df | Union | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
-       "| **Returns** | **Union** | **DataFrame with expected ids and future times missing in `X_df`** |"
+       "| X_df | DFType | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
+       "| **Returns** | **DFType** | **DataFrame with expected ids and future times missing in `X_df`** |"
       ]
      },
      "execution_count": null,
@@ -1896,7 +1892,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L548){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L547){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.forecast_fitted_values\n",
        "\n",
@@ -1914,7 +1910,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L548){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L547){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.forecast_fitted_values\n",
        "\n",
@@ -2337,18 +2333,16 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L619){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L618){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.predict\n",
        "\n",
        ">      MLForecast.predict (h:int,\n",
        ">                          before_predict_callback:Optional[Callable]=None,\n",
-       ">                          after_predict_callback:Optional[Callable]=None, new_d\n",
-       ">                          f:Union[pandas.core.frame.DataFrame,polars.dataframe.\n",
-       ">                          frame.DataFrame,NoneType]=None,\n",
-       ">                          level:Optional[List[Union[int,float]]]=None, X_df:Uni\n",
-       ">                          on[pandas.core.frame.DataFrame,polars.dataframe.frame\n",
-       ">                          .DataFrame,NoneType]=None,\n",
+       ">                          after_predict_callback:Optional[Callable]=None,\n",
+       ">                          new_df:Optional[~DFType]=None,\n",
+       ">                          level:Optional[List[Union[int,float]]]=None,\n",
+       ">                          X_df:Optional[~DFType]=None,\n",
        ">                          ids:Optional[List[str]]=None)\n",
        "\n",
        "*Compute the predictions for the next `h` steps.*\n",
@@ -2358,27 +2352,25 @@
        "| h | int |  | Number of periods to predict. |\n",
        "| before_predict_callback | Optional | None | Function to call on the features before computing the predictions.<br>    This function will take the input dataframe that will be passed to the model for predicting and should return a dataframe with the same structure.<br>    The series identifier is on the index. |\n",
        "| after_predict_callback | Optional | None | Function to call on the predictions before updating the targets.<br>    This function will take a pandas Series with the predictions and should return another one with the same structure.<br>    The series identifier is on the index.    |\n",
-       "| new_df | Union | None | Series data of new observations for which forecasts are to be generated. <br>    This dataframe should have the same structure as the one used to fit the model, including any features and time series data. <br>    If `new_df` is not None, the method will generate forecasts for the new observations. |\n",
+       "| new_df | Optional | None | Series data of new observations for which forecasts are to be generated. <br>    This dataframe should have the same structure as the one used to fit the model, including any features and time series data. <br>    If `new_df` is not None, the method will generate forecasts for the new observations. |\n",
        "| level | Optional | None | Confidence levels between 0 and 100 for prediction intervals. |\n",
-       "| X_df | Union | None | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
+       "| X_df | Optional | None | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
        "| ids | Optional | None | List with subset of ids seen during training for which the forecasts should be computed.         |\n",
-       "| **Returns** | **Union** |  | **Predictions for each serie and timestep, with one column per model.** |"
+       "| **Returns** | **DFType** |  | **Predictions for each serie and timestep, with one column per model.** |"
       ],
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L619){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L618){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.predict\n",
        "\n",
        ">      MLForecast.predict (h:int,\n",
        ">                          before_predict_callback:Optional[Callable]=None,\n",
-       ">                          after_predict_callback:Optional[Callable]=None, new_d\n",
-       ">                          f:Union[pandas.core.frame.DataFrame,polars.dataframe.\n",
-       ">                          frame.DataFrame,NoneType]=None,\n",
-       ">                          level:Optional[List[Union[int,float]]]=None, X_df:Uni\n",
-       ">                          on[pandas.core.frame.DataFrame,polars.dataframe.frame\n",
-       ">                          .DataFrame,NoneType]=None,\n",
+       ">                          after_predict_callback:Optional[Callable]=None,\n",
+       ">                          new_df:Optional[~DFType]=None,\n",
+       ">                          level:Optional[List[Union[int,float]]]=None,\n",
+       ">                          X_df:Optional[~DFType]=None,\n",
        ">                          ids:Optional[List[str]]=None)\n",
        "\n",
        "*Compute the predictions for the next `h` steps.*\n",
@@ -2388,11 +2380,11 @@
        "| h | int |  | Number of periods to predict. |\n",
        "| before_predict_callback | Optional | None | Function to call on the features before computing the predictions.<br>    This function will take the input dataframe that will be passed to the model for predicting and should return a dataframe with the same structure.<br>    The series identifier is on the index. |\n",
        "| after_predict_callback | Optional | None | Function to call on the predictions before updating the targets.<br>    This function will take a pandas Series with the predictions and should return another one with the same structure.<br>    The series identifier is on the index.    |\n",
-       "| new_df | Union | None | Series data of new observations for which forecasts are to be generated. <br>    This dataframe should have the same structure as the one used to fit the model, including any features and time series data. <br>    If `new_df` is not None, the method will generate forecasts for the new observations. |\n",
+       "| new_df | Optional | None | Series data of new observations for which forecasts are to be generated. <br>    This dataframe should have the same structure as the one used to fit the model, including any features and time series data. <br>    If `new_df` is not None, the method will generate forecasts for the new observations. |\n",
        "| level | Optional | None | Confidence levels between 0 and 100 for prediction intervals. |\n",
-       "| X_df | Union | None | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
+       "| X_df | Optional | None | Dataframe with the future exogenous features. Should have the id column and the time column. |\n",
        "| ids | Optional | None | List with subset of ids seen during training for which the forecasts should be computed.         |\n",
-       "| **Returns** | **Union** |  | **Predictions for each serie and timestep, with one column per model.** |"
+       "| **Returns** | **DFType** |  | **Predictions for each serie and timestep, with one column per model.** |"
       ]
      },
      "execution_count": null,
@@ -2947,13 +2939,11 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L206){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L205){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.preprocess\n",
        "\n",
-       ">      MLForecast.preprocess\n",
-       ">                             (df:Union[pandas.core.frame.DataFrame,polars.dataf\n",
-       ">                             rame.frame.DataFrame], id_col:str='unique_id',\n",
+       ">      MLForecast.preprocess (df:~DFType, id_col:str='unique_id',\n",
        ">                             time_col:str='ds', target_col:str='y',\n",
        ">                             static_features:Optional[List[str]]=None,\n",
        ">                             dropna:bool=True, keep_last_n:Optional[int]=None,\n",
@@ -2964,7 +2954,7 @@
        "\n",
        "|    | **Type** | **Default** | **Details** |\n",
        "| -- | -------- | ----------- | ----------- |\n",
-       "| df | Union |  | Series data in long format. |\n",
+       "| df | DFType |  | Series data in long format. |\n",
        "| id_col | str | unique_id | Column that identifies each serie. |\n",
        "| time_col | str | ds | Column that identifies each timestep, its values can be timestamps or integers. |\n",
        "| target_col | str | y | Column that contains the target. |\n",
@@ -2979,13 +2969,11 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L206){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L205){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.preprocess\n",
        "\n",
-       ">      MLForecast.preprocess\n",
-       ">                             (df:Union[pandas.core.frame.DataFrame,polars.dataf\n",
-       ">                             rame.frame.DataFrame], id_col:str='unique_id',\n",
+       ">      MLForecast.preprocess (df:~DFType, id_col:str='unique_id',\n",
        ">                             time_col:str='ds', target_col:str='y',\n",
        ">                             static_features:Optional[List[str]]=None,\n",
        ">                             dropna:bool=True, keep_last_n:Optional[int]=None,\n",
@@ -2996,7 +2984,7 @@
        "\n",
        "|    | **Type** | **Default** | **Details** |\n",
        "| -- | -------- | ----------- | ----------- |\n",
-       "| df | Union |  | Series data in long format. |\n",
+       "| df | DFType |  | Series data in long format. |\n",
        "| id_col | str | unique_id | Column that identifies each serie. |\n",
        "| time_col | str | ds | Column that identifies each timestep, its values can be timestamps or integers. |\n",
        "| target_col | str | y | Column that contains the target. |\n",
@@ -3299,7 +3287,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L262){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L261){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.fit_models\n",
        "\n",
@@ -3318,7 +3306,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L262){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L261){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.fit_models\n",
        "\n",
@@ -3443,15 +3431,13 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L767){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L766){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.cross_validation\n",
        "\n",
-       ">      MLForecast.cross_validation\n",
-       ">                                   (df:Union[pandas.core.frame.DataFrame,polars\n",
-       ">                                   .dataframe.frame.DataFrame], n_windows:int,\n",
-       ">                                   h:int, id_col:str='unique_id',\n",
-       ">                                   time_col:str='ds', target_col:str='y',\n",
+       ">      MLForecast.cross_validation (df:~DFType, n_windows:int, h:int,\n",
+       ">                                   id_col:str='unique_id', time_col:str='ds',\n",
+       ">                                   target_col:str='y',\n",
        ">                                   step_size:Optional[int]=None,\n",
        ">                                   static_features:Optional[List[str]]=None,\n",
        ">                                   dropna:bool=True,\n",
@@ -3472,7 +3458,7 @@
        "\n",
        "|    | **Type** | **Default** | **Details** |\n",
        "| -- | -------- | ----------- | ----------- |\n",
-       "| df | Union |  | Series data in long format. |\n",
+       "| df | DFType |  | Series data in long format. |\n",
        "| n_windows | int |  | Number of windows to evaluate. |\n",
        "| h | int |  | Forecast horizon. |\n",
        "| id_col | str | unique_id | Column that identifies each serie. |\n",
@@ -3491,20 +3477,18 @@
        "| input_size | Optional | None | Maximum training samples per serie in each window. If None, will use an expanding window. |\n",
        "| fitted | bool | False | Store the in-sample predictions. |\n",
        "| as_numpy | bool | False | Cast features to numpy array. |\n",
-       "| **Returns** | **Union** |  | **Predictions for each window with the series id, timestamp, last train date, target value and predictions from each model.** |"
+       "| **Returns** | **DFType** |  | **Predictions for each window with the series id, timestamp, last train date, target value and predictions from each model.** |"
       ],
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L767){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L766){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.cross_validation\n",
        "\n",
-       ">      MLForecast.cross_validation\n",
-       ">                                   (df:Union[pandas.core.frame.DataFrame,polars\n",
-       ">                                   .dataframe.frame.DataFrame], n_windows:int,\n",
-       ">                                   h:int, id_col:str='unique_id',\n",
-       ">                                   time_col:str='ds', target_col:str='y',\n",
+       ">      MLForecast.cross_validation (df:~DFType, n_windows:int, h:int,\n",
+       ">                                   id_col:str='unique_id', time_col:str='ds',\n",
+       ">                                   target_col:str='y',\n",
        ">                                   step_size:Optional[int]=None,\n",
        ">                                   static_features:Optional[List[str]]=None,\n",
        ">                                   dropna:bool=True,\n",
@@ -3525,7 +3509,7 @@
        "\n",
        "|    | **Type** | **Default** | **Details** |\n",
        "| -- | -------- | ----------- | ----------- |\n",
-       "| df | Union |  | Series data in long format. |\n",
+       "| df | DFType |  | Series data in long format. |\n",
        "| n_windows | int |  | Number of windows to evaluate. |\n",
        "| h | int |  | Forecast horizon. |\n",
        "| id_col | str | unique_id | Column that identifies each serie. |\n",
@@ -3544,7 +3528,7 @@
        "| input_size | Optional | None | Maximum training samples per serie in each window. If None, will use an expanding window. |\n",
        "| fitted | bool | False | Store the in-sample predictions. |\n",
        "| as_numpy | bool | False | Cast features to numpy array. |\n",
-       "| **Returns** | **Union** |  | **Predictions for each window with the series id, timestamp, last train date, target value and predictions from each model.** |"
+       "| **Returns** | **DFType** |  | **Predictions for each window with the series id, timestamp, last train date, target value and predictions from each model.** |"
       ]
      },
      "execution_count": null,
@@ -4286,6 +4270,7 @@
     "    h=horizon,\n",
     "    step_size=horizon,\n",
     "    input_size=input_size,\n",
+    "    keep_last_n=input_size,\n",
     ")\n",
     "series_lengths = np.diff(fcst.ts.ga.indptr)\n",
     "unique_lengths = np.unique(series_lengths)\n",
@@ -4437,7 +4422,7 @@
       "text/markdown": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L192){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L191){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.from_cv\n",
        "\n",
@@ -4446,7 +4431,7 @@
       "text/plain": [
        "---\n",
        "\n",
-       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L192){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
+       "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/forecast.py#L191){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
        "\n",
        "### MLForecast.from_cv\n",
        "\n",
@@ -4480,7 +4465,6 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "[LightGBM] [Info] Start training from score 0.084340\n",
       "[10] mape: 0.118569\n",
       "[20] mape: 0.111506\n",
       "[30] mape: 0.107314\n",
diff --git a/nbs/lag_transforms.ipynb b/nbs/lag_transforms.ipynb
index 9caac5ee..2eb96f58 100644
--- a/nbs/lag_transforms.ipynb
+++ b/nbs/lag_transforms.ipynb
@@ -113,7 +113,15 @@
     "        out._core_tfm = transforms[0]._core_tfm.stack(\n",
     "            [tfm._core_tfm for tfm in transforms]\n",
     "        )\n",
-    "        return out"
+    "        return out\n",
+    "\n",
+    "    @property\n",
+    "    def _lag(self):\n",
+    "        return self._core_tfm.lag - 1\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return -1"
    ]
   },
   {
@@ -125,6 +133,7 @@
    "source": [
     "#| exporti\n",
     "class Lag(_BaseLagTransform):\n",
+    "    \n",
     "    def __init__(self, lag: int):\n",
     "        self.lag = lag\n",
     "        self._core_tfm = core_tfms.Lag(lag=lag)\n",
@@ -136,7 +145,11 @@
     "        return f'lag{lag}'\n",
     "\n",
     "    def __eq__(self, other):\n",
-    "        return isinstance(other, Lag) and self.lag == other.lag"
+    "        return isinstance(other, Lag) and self.lag == other.lag\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return self.lag"
    ]
   },
   {
@@ -160,7 +173,11 @@
     "            If `None`, will be set to `window_size`.\n",
     "        \"\"\"\n",
     "        self.window_size = window_size\n",
-    "        self.min_samples = min_samples"
+    "        self.min_samples = min_samples\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return self._lag + self.window_size"
    ]
   },
   {
@@ -248,7 +265,11 @@
     "        \"\"\"        \n",
     "        self.season_length = season_length\n",
     "        self.window_size = window_size\n",
-    "        self.min_samples = min_samples"
+    "        self.min_samples = min_samples\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return self._lag + self.season_length * self.window_size"
    ]
   },
   {
@@ -311,7 +332,11 @@
     "class _ExpandingBase(_BaseLagTransform):\n",
     "    \"\"\"Expanding statistic\"\"\"\n",
     "    def __init__(self):\n",
-    "        ..."
+    "        ...\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return 1"
    ]
   },
   {
@@ -336,7 +361,11 @@
     "\n",
     "class ExpandingQuantile(_ExpandingBase):\n",
     "    def __init__(self, p: float):\n",
-    "        self.p = p"
+    "        self.p = p\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return -1"
    ]
   },
   {
@@ -378,7 +407,11 @@
     "    alpha : float\n",
     "        Smoothing factor.\"\"\"\n",
     "    def __init__(self, alpha: float):\n",
-    "        self.alpha = alpha"
+    "        self.alpha = alpha\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return 1"
    ]
   },
   {
@@ -429,8 +462,13 @@
     "        return self.tfm._get_name(lag + self.n)\n",
     "\n",
     "    def _set_core_tfm(self, lag: int) -> 'Offset':\n",
-    "        self._core_tfm = clone(self.tfm)._set_core_tfm(lag + self.n)\n",
-    "        return self"
+    "        self.tfm = clone(self.tfm)._set_core_tfm(lag + self.n)\n",
+    "        self._core_tfm = self.tfm._core_tfm\n",
+    "        return self\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self) -> int:\n",
+    "        return self.tfm.update_samples + self.n"
    ]
   },
   {
@@ -488,7 +526,11 @@
     "        return self.operator(self.tfm1.transform(ga), self.tfm2.transform(ga))\n",
     "\n",
     "    def update(self, ga: CoreGroupedArray) -> np.ndarray:\n",
-    "        return self.operator(self.tfm1.update(ga), self.tfm2.update(ga))"
+    "        return self.operator(self.tfm1.update(ga), self.tfm2.update(ga))\n",
+    "\n",
+    "    @property\n",
+    "    def update_samples(self):\n",
+    "        return max(self.tfm1.update_samples, self.tfm2.update_samples)"
    ]
   },
   {
@@ -499,7 +541,9 @@
    "outputs": [],
    "source": [
     "#| hide\n",
-    "import operator"
+    "import operator\n",
+    "\n",
+    "from mlforecast.grouped_array import GroupedArray as MLGroupedArray"
    ]
   },
   {
@@ -551,7 +595,14 @@
     "    tfm._set_core_tfm(1)\n",
     "    tfm._get_name(1)\n",
     "    tfm.transform(ga)\n",
-    "    tfm.update(ga)"
+    "    updates = tfm.update(ga)\n",
+    "    upd_samples = tfm.update_samples\n",
+    "    if upd_samples > -1:\n",
+    "        sliced_ga = MLGroupedArray(ga.data, ga.indptr).take_from_groups(slice(-upd_samples, None))\n",
+    "        ga2 = CoreGroupedArray(sliced_ga.data, sliced_ga.indptr)\n",
+    "        tfm.transform(ga)  # to reset state\n",
+    "        updates2 = tfm.update(ga2)\n",
+    "        np.testing.assert_allclose(updates, updates2)"
    ]
   }
  ],