Add Cyclic Boosting Machines (CBM) integration

environment.yml

@@ -18,6 +18,7 @@ dependencies:
   - window-ops
   - xgboost
   - pip:
+    - cyclicbm
     - datasetsforecast
     - fugue[ray]
     - lightgbm_ray

mlforecast/core.py

@@ -405,10 +405,11 @@ def fit_transform(
     def _update_y(self, new: np.ndarray) -> None:
         """Appends the elements of `new` to every time serie.
 
-        These values are used to update the transformations and are stored as predictions.
-        """
+        These values are used to update the transformations and are stored as predictions."""
         if not hasattr(self, "y_pred"):
             self.y_pred = []
+        if new.ndim > 1:
+            new = new.flatten()
         self.y_pred.append(new)
         new_arr = np.asarray(new)
         self.ga = self.ga.append(new_arr)

nbs/core.ipynb

@@ -856,6 +856,8 @@
     "        These values are used to update the transformations and are stored as predictions.\"\"\"\n",
     "        if not hasattr(self, 'y_pred'):\n",
     "            self.y_pred = []\n",
+    "        if new.ndim > 1:\n",
+    "            new = new.flatten()\n",
     "        self.y_pred.append(new)\n",
     "        new_arr = np.asarray(new)\n",
     "        self.ga = self.ga.append(new_arr)     \n",

nbs/forecast.ipynb

@@ -1661,6 +1661,55 @@
     ")"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "07a06d35-77c4-417f-83fa-7057167a5f71",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| hide\n",
+    "import cbm"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8c2d0009-1a41-4dc3-868f-3cdab0265ef9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| hide\n",
+    "# cyclic boosting machine integration\n",
+    "fcst_test = MLForecast(\n",
+    "    models=cbm.CBM(),\n",
+    "    lags=[1],\n",
+    "    num_threads=1,\n",
+    "    freq='D'\n",
+    ")\n",
+    "df_test = generate_daily_series(1, n_static_features=2, static_as_categorical=True)\n",
+    "fcst_test.fit(\n",
+    "    df_test,\n",
+    "    static_features=['static_0', 'static_1'],\n",
+    "    prediction_intervals=PredictionIntervals(),\n",
+    ")\n",
+    "pred_test = fcst_test.predict(12)\n",
+    "pred_int_test = fcst_test.predict(12, level=[80, 90])\n",
+    "# test same structure\n",
+    "test_eq(\n",
+    "    pred_test,\n",
+    "    pred_int_test[pred_test.columns]\n",
+    ")\n",
+    "# test monotonicity of intervals\n",
+    "test_eq(\n",
+    "    pred_int_test.filter(regex='lo|hi').apply(\n",
+    "        lambda x: x.is_monotonic_increasing,\n",
+    "        axis=1\n",
+    "    ).sum(),\n",
+    "    len(pred_int_test)\n",
+    ")"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "505aadfd-076d-42e6-8103-aeef4eef61b7",

settings.ini

@@ -17,7 +17,7 @@ license = apache2
 status = 3
 requirements = numba pandas scikit-learn window-ops
 distributed_requirements = dask[complete] fugue[ray]  pyspark lightgbm_ray xgboost_ray
-dev_requirements = black datasetsforecast flake8 lightgbm matplotlib mypy nbdev pylint statsforecast xgboost
+dev_requirements = black cyclicbm datasetsforecast flake8 lightgbm matplotlib mypy nbdev pylint statsforecast xgboost
 nbs_path = nbs
 doc_path = _docs
 recursive = True