Add predict_columns parameter to cross validation (#2486)

python/prophet/diagnostics.py

@@ -58,7 +58,7 @@ def generate_cutoffs(df, horizon, initial, period):
     return list(reversed(result))
 
 
-def cross_validation(model, horizon, period=None, initial=None, parallel=None, cutoffs=None, disable_tqdm=False):
+def cross_validation(model, horizon, period=None, initial=None, parallel=None, cutoffs=None, disable_tqdm=False, extra_output_columns=None):
     """Cross-Validation for time series.
 
     Computes forecasts from historical cutoff points, which user can input.
@@ -82,8 +82,6 @@ def cross_validation(model, horizon, period=None, initial=None, parallel=None, c
         cross validation. If not provided, they are generated as described
         above.
     parallel : {None, 'processes', 'threads', 'dask', object}
-    disable_tqdm: if True it disables the progress bar that would otherwise show up when parallel=None
-
         How to parallelize the forecast computation. By default no parallelism
         is used.
 
@@ -110,6 +108,10 @@ def cross_validation(model, horizon, period=None, initial=None, parallel=None, c
                      ]
                      return results
                      
+    disable_tqdm: if True it disables the progress bar that would otherwise show up when parallel=None
+    extra_output_columns: A String or List of Strings e.g. 'trend' or ['trend'].
+         Additional columns to 'yhat' and 'ds' to be returned in output.
+
     Returns
     -------
     A pd.DataFrame with the forecast, actual value and cutoff.
@@ -120,11 +122,16 @@ def cross_validation(model, horizon, period=None, initial=None, parallel=None, c
     
     df = model.history.copy().reset_index(drop=True)
     horizon = pd.Timedelta(horizon)
-
     predict_columns = ['ds', 'yhat']
+        
     if model.uncertainty_samples:
         predict_columns.extend(['yhat_lower', 'yhat_upper'])
 
+    if extra_output_columns is not None:
+        if isinstance(extra_output_columns, str):
+            extra_output_columns = [extra_output_columns]
+        predict_columns.extend([c for c in extra_output_columns if c not in predict_columns])
+        
     # Identify largest seasonality period
     period_max = 0.
     for s in model.seasonalities.values():

python/prophet/tests/test_diagnostics.py

@@ -102,6 +102,19 @@ class TestCrossValidation:
             # check single forecast function called expected number of times
             assert n_calls == forecasts
     
+    @pytest.mark.parametrize("extra_output_columns", ["trend", ["trend"]])
+    def test_check_extra_output_columns_cross_validation(self, ts_short, backend, extra_output_columns):
+        m = Prophet(stan_backend=backend)
+        m.fit(ts_short)
+        df_cv = diagnostics.cross_validation(
+            m,
+            horizon="1 days",
+            period="1 days",
+            initial="140 days",
+            extra_output_columns=extra_output_columns
+        )
+        assert "trend" in df_cv.columns
+
     @pytest.mark.parametrize("growth", ["logistic", "flat"])
     def test_cross_validation_logistic_or_flat_growth(self, growth, ts_short, backend):
         df = ts_short.copy()