Extend unit tests

python/prophet/tests/test_diagnostics.py

@@ -12,6 +12,7 @@
 
 from prophet import Prophet, diagnostics
 
+import logging
 
 @pytest.fixture(scope="module")
 def ts_short(daily_univariate_ts):
@@ -401,3 +402,49 @@ def test_prophet_copy_custom(self, data, backend):
         assert (changepoints == m2.changepoints).all()
         assert "custom" in m2.seasonalities
         assert "binary_feature" in m2.extra_regressors
+
+
+    def test_cross_validation_dask_import_error(self, ts_short, backend, monkeypatch):
+        m = Prophet(stan_backend=backend)
+        m.fit(ts_short)
+
+        # Simulate dask not being installed
+        import sys
+        if 'dask' in sys.modules:
+            del sys.modules['dask']
+
+        with pytest.raises(ImportError, match="parallel='dask' requires the optional dependency dask."):
+            diagnostics.cross_validation(m, horizon="4 days", parallel="dask")
+
+
+    def test_cross_validation_min_cutoff_value_error(self, ts_short, backend):
+        m = Prophet(stan_backend=backend)
+        m.fit(ts_short)
+        cutoffs = [ts_short['ds'].min(), pd.Timestamp("2012-07-01")]
+        with pytest.raises(ValueError, match="Minimum cutoff value is not strictly greater than min date in history"):
+            diagnostics.cross_validation(m, horizon="4 days", cutoffs=cutoffs)
+
+
+    def test_cross_validation_initial_window_warning(self, ts_short, backend, caplog):
+        m = Prophet(stan_backend=backend)
+        m.add_seasonality(name="monthly", period=30.5, fourier_order=5)
+        m.fit(ts_short)
+        with caplog.at_level(logging.WARNING):
+            diagnostics.cross_validation(
+                m, horizon="4 days", period="10 days", initial="20 days"
+            )
+        assert any("Seasonality has period of" in message for message in caplog.messages)
+
+
+    def test_cross_validation_max_cutoff_value_error(self, ts_short, backend):
+        m = Prophet(stan_backend=backend)
+        m.fit(ts_short)
+        cutoffs = [pd.Timestamp("2012-12-31")]
+        with pytest.raises(ValueError, match="Maximum cutoff value is greater than end date minus horizon"):
+            diagnostics.cross_validation(m, horizon="4 days", cutoffs=cutoffs)
+
+
+    def test_cross_validation_unfit_model_exception(self):
+        m = Prophet()
+        with pytest.raises(Exception, match="Model has not been fit"):
+            diagnostics.cross_validation(m, horizon="4 days")

python/prophet/tests/test_prophet.py

@@ -982,3 +982,47 @@ def test_sampling_warm_start(self, daily_univariate_ts, backend):
             daily_univariate_ts.iloc[:510], init=warm_start_params(m), show_progress=False
         )
         assert m2.params["delta"].shape == (200, 25)
+
+
+    def test_invalid_scaling_parameter(self, daily_univariate_ts, backend):
+        with pytest.raises(ValueError, match="scaling must be one of 'absmax' or 'minmax'"):
+            Prophet(scaling="invalid", stan_backend=backend).fit(daily_univariate_ts)
+
+
+    def test_holidays_missing_ds_column(self, daily_univariate_ts, backend):
+        holidays = pd.DataFrame({
+            "holiday": ["xmas"]
+        })
+        with pytest.raises(ValueError, match='holidays must be a DataFrame with "ds" and "holiday" columns.'):
+            Prophet(holidays=holidays, stan_backend=backend).fit(daily_univariate_ts)
+
+
+    def test_holidays_upper_window_less_than_zero(self, daily_univariate_ts, backend):
+        holidays = pd.DataFrame({
+            "ds": pd.to_datetime(["2016-12-25"]),
+            "holiday": ["xmas"],
+            "lower_window": [-1],
+            "upper_window": [-2]
+        })
+        with pytest.raises(ValueError, match="Holiday upper_window should be >= 0"):
+            Prophet(holidays=holidays, stan_backend=backend).fit(daily_univariate_ts)
+
+
+    def test_holidays_lower_window_greater_than_zero(self, daily_univariate_ts, backend):
+        holidays = pd.DataFrame({
+            "ds": pd.to_datetime(["2016-12-25"]),
+            "holiday": ["xmas"],
+            "lower_window": [1],
+            "upper_window": [0]
+        })
+        with pytest.raises(ValueError, match="Holiday lower_window should be <= 0"):
+            Prophet(holidays=holidays, stan_backend=backend).fit(daily_univariate_ts)
+
+
+    def test_holidays_with_nan(self, daily_univariate_ts, backend):
+        holidays = pd.DataFrame({
+            "ds": pd.to_datetime(["2016-12-25", None]),
+            "holiday": ["xmas", "xmas"]
+        })
+        with pytest.raises(ValueError, match="Found a NaN in holidays dataframe."):
+            Prophet(holidays=holidays, stan_backend=backend).fit(daily_univariate_ts)

python/prophet/tests/test_serialize.py

@@ -140,3 +140,23 @@ def test_backwards_compatibility(self):
             future = m.make_future_dataframe(10)
             fcst = m.predict(future)
             assert fcst["yhat"].values[-1] == pytest.approx(pred_val)
+
+
+    def test_fit_kwargs_conversion(self, daily_univariate_ts, backend):
+        from prophet.serialize import model_to_dict
+        m = Prophet(stan_backend=backend)
+        df = daily_univariate_ts.head(daily_univariate_ts.shape[0] - 30)
+        m.fit(df)
+        m.fit_kwargs['init'] = {'param1': np.array([1.0, 2.0]), 'param2': np.float64(3.0)}
+
+        model_dict = model_to_dict(m)
+
+        assert model_dict['fit_kwargs']['init']['param1'] == [1.0, 2.0]
+        assert model_dict['fit_kwargs']['init']['param2'] == 3.0
+
+
+    def test_model_to_dict_unfitted_model(self):
+        from prophet.serialize import model_to_dict
+        m = Prophet()
+        with pytest.raises(ValueError, match="This can only be used to serialize models that have already been fit."):
+            model_to_dict(m)

python/prophet/tests/test_utilities.py

@@ -8,6 +8,7 @@
 from prophet import Prophet
 from prophet.utilities import regressor_coefficients
 
+from prophet.utilities import warm_start_params
 
 class TestUtilities:
     def test_regressor_coefficients(self, daily_univariate_ts, backend):
@@ -25,3 +26,37 @@ def test_regressor_coefficients(self, daily_univariate_ts, backend):
         # No MCMC sampling, so lower and upper should be the same as mean
         assert np.array_equal(coefs["coef_lower"].values, coefs["coef"].values)
         assert np.array_equal(coefs["coef_upper"].values, coefs["coef"].values)
+
+    def test_warm_start_params_with_mcmc(self, daily_univariate_ts, backend):
+        m = Prophet(stan_backend=backend, mcmc_samples=10)
+        df = daily_univariate_ts.copy()
+        np.random.seed(123)
+        df["regr1"] = np.random.normal(size=df.shape[0])
+        m.add_regressor("regr1", mode="additive")
+        m.fit(df)
+
+        params = warm_start_params(m)
+        assert isinstance(params, dict)
+        assert "k" in params
+        assert "m" in params
+        assert "sigma_obs" in params
+        assert "delta" in params
+        assert "beta" in params
+
+
+    def test_warm_start_params_no_mcmc(self, daily_univariate_ts, backend):
+        m = Prophet(stan_backend=backend, mcmc_samples=0)
+        df = daily_univariate_ts.copy()
+        np.random.seed(123)
+        df["regr1"] = np.random.normal(size=df.shape[0])
+        m.add_regressor("regr1", mode="additive")
+        m.fit(df)
+
+        params = warm_start_params(m)
+        assert isinstance(params, dict)
+        assert "k" in params
+        assert "m" in params
+        assert "sigma_obs" in params
+        assert "delta" in params
+        assert "beta" in params
+