fix: solve problem with duplicated dates (#190)

Nixtla · Dec 9, 2023 · 5897e21 · 5897e21
2 parents 990090e + cc5c0e8
commit 5897e21
Show file tree

Hide file tree

Showing 4 changed files with 29 additions and 9 deletions.
diff --git a/environment.yml b/environment.yml
@@ -2,10 +2,11 @@ name: nixtlats
 channels:
   - conda-forge
 dependencies:
-  - dask<2023.1.1
+  - dask
   - jupyterlab
   - pandas
   - plotly
+  - prophet
   - pyspark>=3.3
   - pip:
     - black
@@ -17,9 +18,8 @@ dependencies:
     - statsforecast
     - utilsforecast>=0.0.13
     - requests
-    - duckdb<0.8
-    - fugue[ray]
-    - ray<2.4
+    - fugue[ray]>=0.8.7
+    - ray[serve-grpc]
     - fire
     - tabulate
     - tenacity
diff --git a/nbs/timegpt.ipynb b/nbs/timegpt.ipynb
@@ -276,7 +276,7 @@
     "        if self.freq is None or self.freq in special_freqs:\n",
     "            unique_id = df.iloc[0]['unique_id']\n",
     "            df_id = df.query('unique_id == @unique_id')\n",
-    "            inferred_freq = pd.infer_freq(df_id['ds'])\n",
+    "            inferred_freq = pd.infer_freq(df_id['ds'].sort_values())\n",
     "            if inferred_freq is None:\n",
     "                raise Exception(\n",
     "                    'Could not infer frequency of ds column. This could be due to '\n",
@@ -345,7 +345,7 @@
     "            future_dates = X_df['ds'].unique().tolist()\n",
     "        else:\n",
     "            future_dates = []\n",
-    "        dates = pd.DatetimeIndex(train_dates + future_dates)\n",
+    "        dates = pd.DatetimeIndex(np.unique(train_dates + future_dates).tolist())\n",
     "        date_features_df = pd.DataFrame({'ds': dates})\n",
     "        for feature in self.date_features:\n",
     "            feat_df = self.compute_date_feature(dates, feature)\n",
@@ -1546,6 +1546,26 @@
     "from utilsforecast.data import generate_series"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| hide\n",
+    "# test date_features with multiple series\n",
+    "# and different ends\n",
+    "test_series = generate_series(n_series=2, min_length=5, max_length=20)\n",
+    "h = 12\n",
+    "fcst_test_series = timegpt.forecast(test_series, h=12, date_features=['dayofweek'])\n",
+    "uids = test_series['unique_id']\n",
+    "for uid in uids:\n",
+    "    test_eq(\n",
+    "        fcst_test_series.query('unique_id == @uid')['ds'].values,\n",
+    "        pd.date_range(periods=h + 1, start=test_series.query('unique_id == @uid')['ds'].max())[1:].astype(str),\n",
+    "    )"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,

diff --git a/nixtlats/timegpt.py b/nixtlats/timegpt.py
@@ -218,7 +218,7 @@ def infer_freq(self, df: pd.DataFrame):
         if self.freq is None or self.freq in special_freqs:
             unique_id = df.iloc[0]["unique_id"]
             df_id = df.query("unique_id == @unique_id")
-            inferred_freq = pd.infer_freq(df_id["ds"])
+            inferred_freq = pd.infer_freq(df_id["ds"].sort_values())
             if inferred_freq is None:
                 raise Exception(
                     "Could not infer frequency of ds column. This could be due to "
@@ -295,7 +295,7 @@ def add_date_features(
             future_dates = X_df["ds"].unique().tolist()
         else:
             future_dates = []
-        dates = pd.DatetimeIndex(train_dates + future_dates)
+        dates = pd.DatetimeIndex(np.unique(train_dates + future_dates).tolist())
         date_features_df = pd.DataFrame({"ds": dates})
         for feature in self.date_features:
             feat_df = self.compute_date_feature(dates, feature)

diff --git a/setup.py b/setup.py
@@ -4,7 +4,7 @@
     long_description = fh.read()
 
 dev = ["black", "nbdev", "plotly", "python-dotenv", "statsforecast"]
-distributed = ["dask", "fugue[ray]", "pyspark"]
+distributed = ["dask", "fugue[ray]>=0.8.7", "pyspark", "ray[serve-grpc]"]
 plotting = ["utilsforecast[plotting]>=0.0.5"]
 date_extras = ["holidays"]