test fixes for fbcode/kats

Summary:
Some PSS1-pinned directories have issues that trace back to fbcode/kats, so trying to upgrade it first. This diff makes backwards-compatible fixes to several test failures seen on D63678019:

**bocpd_model.py**
- [HoltWintersResults.slope deprecated in favor of .trend](https://www.statsmodels.org/v0.12.2/generated/statsmodels.tsa.holtwinters.HoltWintersResults.slope.html#statsmodels.tsa.holtwinters.HoltWintersResults.slope)

**special_events_base.py**
- assuming that self.original_fcst is supposed to be a copy, ignore chained_assignment warning

**test_dtw.py**
- freq field on pandas.Timestamp is deprecated and doesn't do anything anyways

**decomposition.py**
- first argument is expected to be 1D, achievable with .squeeze()

Changes copied from proof-by-accident D58854294:
- changepoint_evaluator.py
- stl_intraday_model.py
- test_feature_engineering.py
- feature_engineering.py

everything else is .append -> pandas.concat

Reviewed By: igorsugak

Differential Revision: D63737460

fbshipit-source-id: 66b3cebc7d3a9a912de6f6674b4c587b462217b5

kats/detectors/bocpd_model.py

@@ -223,7 +223,7 @@ def _holt_winter_fit(
         )
 
         level_arr = fit1.level
-        trend_arr = fit1.slope
+        trend_arr = fit1.trend
         fit_arr = [x + y for x, y in zip(level_arr, trend_arr)]
         fit_diff = np.diff(fit_arr)
         fit_diff = np.concatenate(([fit_diff[0]], fit_diff))

kats/detectors/detector_consts.py

@@ -577,27 +577,29 @@ def update(
     def _update_ts_slice(
         self, ts: TimeSeriesData, time: datetime, value: Union[float, ArrayLike]
     ) -> TimeSeriesData:
-        time = ts.time.iloc[1:].append(pd.Series(time, copy=False))
-        time.reset_index(drop=True, inplace=True)
+        time_df = pd.concat([ts.time.iloc[1:], pd.Series(time, copy=False)])
+        time_df.reset_index(drop=True, inplace=True)
         if self.num_series == 1:
-            value = ts.value.iloc[1:].append(pd.Series(value, copy=False))
-            value.reset_index(drop=True, inplace=True)
-            return TimeSeriesData(time=time, value=value)
+            value_df = pd.concat([ts.value.iloc[1:], pd.Series(value, copy=False)])
+            value_df.reset_index(drop=True, inplace=True)
+            # pyre-fixme[6]: For 1st argument expected `Union[None, DatetimeIndex,
+            #  Series]` but got `DataFrame`.
+            return TimeSeriesData(time=time_df, value=value_df)
         else:
             if isinstance(value, float):
                 raise ValueError(
                     f"num_series = {self.num_series} so value should have type ArrayLike."
                 )
             value_dict = {}
             for i, value_col in enumerate(self.key_mapping):
-                value_dict[value_col] = (
-                    ts.value[value_col].iloc[1:].append(pd.Series(value[i], copy=False))
+                value_dict[value_col] = pd.concat(
+                    [ts.value[value_col].iloc[1:], pd.Series(value[i], copy=False)]
                 )
                 value_dict[value_col].reset_index(drop=True, inplace=True)
             return TimeSeriesData(
                 pd.DataFrame(
                     {
-                        **{"time": time},
+                        **{"time": time_df},
                         **{
                             value_col: value_dict[value_col]
                             for value_col in self.key_mapping

kats/detectors/distribution_distance_model.py

@@ -285,7 +285,7 @@ def fit_predict(
                     total_data_df_group0.loc[start_time_index:],
                     total_data_df.loc[start_time_index:],
                 ],
-                1,
+                axis=1,
                 copy=False,
             )
             scores = total_df.apply(self._js_div_func, axis=1)

kats/detectors/trend_mk.py

@@ -442,8 +442,8 @@ def detector(
             ts_deseas = self._remove_seasonality(ts, freq=self.freq)
             ts_smoothed = self._smoothing(ts_deseas)  # smoothing
             # append MK statistics to MK_statistics dataframe
-            MK_statistics = MK_statistics.append(
-                self.runDetector(ts=ts_smoothed),
+            MK_statistics = pd.concat(
+                [MK_statistics, pd.DataFrame([self.runDetector(ts=ts_smoothed)])],
                 ignore_index=True,
             )
 
@@ -458,8 +458,8 @@ def detector(
                 # look back window_size day for trend detection
                 ts_tmp = ts_smoothed.loc[:t, :]
                 # append MK statistics to MK_statistics dataframe
-                MK_statistics = MK_statistics.append(
-                    self.runDetector(ts=ts_tmp),
+                MK_statistics = pd.concat(
+                    [MK_statistics, pd.DataFrame([self.runDetector(ts=ts_tmp)])],
                     ignore_index=True,
                 )
 

kats/models/globalmodel/utils.py

@@ -1686,8 +1686,13 @@ def pad_ts(ts: TimeSeriesData, n: int, freq: pd.Timedelta) -> TimeSeriesData:
     val_col = [col for col in df.columns if col != time_col][0]
     pad_val = df[val_col].values[df[val_col].first_valid_index()]
     # add the padding value
-    df = df.append(
-        {time_col: df[time_col].iloc[0] - n * freq, val_col: pad_val},
+    df = pd.concat(
+        [
+            df,
+            pd.DataFrame(
+                [{time_col: df[time_col].iloc[0] - n * freq, val_col: pad_val}]
+            ),
+        ],
         ignore_index=True,
     )
     return TimeSeriesData(df)

kats/tests/data/test_data_validation.py

@@ -23,7 +23,7 @@ def test_data_validation(self) -> None:
             [["1900-01-01", 2], ["2020-01-01", 2]], columns=["time", "y"]
         )
         DATA = self.TSData.to_dataframe()
-        data_with_extra_point = DATA.copy().append(extra_point)
+        data_with_extra_point = pd.concat([DATA.copy(), extra_point])
 
         tsData_with_missing_point = TimeSeriesData(data_with_extra_point)
 

kats/tests/detectors/test_distribution_distance_model.py

@@ -45,7 +45,7 @@ def generate_multi_ts_data(
 
     multi_ts_df = pd.DataFrame(multi_ts_val)
     multi_ts_df.columns = ["val_" + str(i) for i in range(10)]
-    multi_ts_df = pd.concat([pd.Series(ts_time, name="time"), multi_ts_df], 1)
+    multi_ts_df = pd.concat([pd.Series(ts_time, name="time"), multi_ts_df], axis=1)
     # pyre-fixme[6]: For 1st argument expected `Optional[DataFrame]` but got
     #  `Union[DataFrame, Series]`.
     ts = TimeSeriesData(multi_ts_df)
@@ -134,7 +134,7 @@ def generate_irregular_granularity_data(
         start="2020-01-01", freq=str(granularities_sec) + "s", periods=length
     )[np.random.choice(list(range(length)), n, replace=False)]
 
-    multi_ts_df = pd.concat([pd.Series(ts_time, name="time"), multi_ts_df], 1)
+    multi_ts_df = pd.concat([pd.Series(ts_time, name="time"), multi_ts_df], axis=1)
     # pyre-fixme[6]: For 1st argument expected `Optional[DataFrame]` but got
     #  `Union[DataFrame, Series]`.
     ts = TimeSeriesData(multi_ts_df)

kats/tests/detectors/test_dtw.py

@@ -45,8 +45,8 @@ def test_clear_spike_in_zero_data_yields_cp(self) -> None:
 
         expected_result = [
             DTWCPDChangePoint(
-                start_time=pd.Timestamp("2021-03-02 00:00:00", freq="D"),
-                end_time=pd.Timestamp("2021-04-20 00:00:00", freq="D"),
+                start_time=pd.Timestamp("2021-03-02 00:00:00"),
+                end_time=pd.Timestamp("2021-04-20 00:00:00"),
                 confidence=1e9,
                 ts_name="ts3",
             )
@@ -99,8 +99,8 @@ def test_two_similar_spikes_in_zero_data_yields_cp(self) -> None:
 
         expected_result = [
             DTWCPDChangePoint(
-                start_time=pd.Timestamp("2021-03-02 00:00:00", freq="D"),
-                end_time=pd.Timestamp("2021-04-20 00:00:00", freq="D"),
+                start_time=pd.Timestamp("2021-03-02 00:00:00"),
+                end_time=pd.Timestamp("2021-04-20 00:00:00"),
                 confidence=mock.ANY,
                 ts_name="ts3",
             )

kats/tests/detectors/test_outlier.py

@@ -131,7 +131,9 @@ class MultivariateVARDetectorTest(TestCase):
     def setUp(self) -> None:
         DATA_multi = load_data("multivariate_anomaly_simulated_data.csv")
         self.TSData_multi = TimeSeriesData(DATA_multi)
-        DATA_multi2 = pd.concat([DATA_multi.iloc[:10, :], DATA_multi.iloc[12:, :]], 0)
+        DATA_multi2 = pd.concat(
+            [DATA_multi.iloc[:10, :], DATA_multi.iloc[12:, :]], axis=0
+        )
         self.TSData_multi2 = TimeSeriesData(DATA_multi2)
         DATA_multi3 = pd.merge(
             DATA_multi, DATA_multi, how="inner", on="time", suffixes=("_1", "_2")

kats/tests/test_consts.py

@@ -1170,15 +1170,15 @@ def setUp(self) -> None:
         transformed_df_date.ds = transformed_df_date.ds.apply(
             lambda x: x + relativedelta(years=NUM_YEARS_OFFSET)
         )
-        transformed_df_date_concat = self.AIR_DF.append(
-            transformed_df_date, ignore_index=True
+        transformed_df_date_concat = pd.concat(
+            [self.AIR_DF, transformed_df_date], ignore_index=True
         )
         transformed_df_date_double = self.AIR_DF_DATETIME.copy(deep=True)
         transformed_df_date_double.ds = transformed_df_date.ds.apply(
             lambda x: x + relativedelta(years=NUM_YEARS_OFFSET * 2)
         )
-        transformed_df_date_concat_double = self.AIR_DF.append(
-            transformed_df_date_double, ignore_index=True
+        transformed_df_date_concat_double = pd.concat(
+            [self.AIR_DF, transformed_df_date_double], ignore_index=True
         )
         # DataFrames with value offset
         transformed_df_value = self.AIR_DF.copy(deep=True)
@@ -1195,21 +1195,21 @@ def setUp(self) -> None:
         transformed_df_date_multi[VALUE_COL_NAME + "_1"] = (
             transformed_df_date_multi.y * 2
         )
-        transformed_df_date_concat_multi = self.MULTIVAR_AIR_DF.append(
-            transformed_df_date_multi, ignore_index=True
+        transformed_df_date_concat_multi = pd.concat(
+            [self.MULTIVAR_AIR_DF, transformed_df_date_multi], ignore_index=True
         )
-        transformed_df_date_concat_mixed = self.MULTIVAR_AIR_DF_DATETIME.append(
-            transformed_df_date
+        transformed_df_date_concat_mixed = pd.concat(
+            [self.MULTIVAR_AIR_DF_DATETIME, transformed_df_date]
         )
         transformed_df_date_double_multi = transformed_df_date_double.copy(deep=True)
         transformed_df_date_double_multi[VALUE_COL_NAME + "_1"] = (
             transformed_df_date_double_multi.y * 2
         )
-        transformed_df_date_concat_double_multi = self.MULTIVAR_AIR_DF.append(
-            transformed_df_date_double_multi, ignore_index=True
+        transformed_df_date_concat_double_multi = pd.concat(
+            [self.MULTIVAR_AIR_DF, transformed_df_date_double_multi], ignore_index=True
         )
-        transformed_df_date_concat_double_mixed = self.MULTIVAR_AIR_DF_DATETIME.append(
-            transformed_df_date_double
+        transformed_df_date_concat_double_mixed = pd.concat(
+            [self.MULTIVAR_AIR_DF_DATETIME, transformed_df_date_double]
         )
         # DataFrame with value offset (multivariate)
         transformed_df_value_none_multi = self.MULTIVAR_AIR_DF.copy(deep=True)

kats/tests/utils/test_feature_engineering.py

@@ -78,7 +78,7 @@ def test_datetime_features(self) -> None:
             index=dates,
         )
         result = fe.datetime_features(pd.Series(values, index=dates, name="val"))
-        assert_frame_equal(expected, result)
+        assert_frame_equal(expected, result, check_like=True, check_dtype=False)
 
     def test_date_features(self) -> None:
         dates = pd.date_range("2021-01-22", "2021-01-31", tz="US/Pacific").tolist()
@@ -109,7 +109,7 @@ def test_date_features(self) -> None:
             index=dates,
         )
         result = fe.date_features(pd.Series(values, index=dates, name="val"))
-        assert_frame_equal(expected, result)
+        assert_frame_equal(expected, result, check_like=True, check_dtype=False)
 
     def test_time_features(self) -> None:
         dates = pd.date_range("2021-01-22", "2021-01-31", tz="US/Pacific").tolist()
@@ -134,7 +134,7 @@ def test_time_features(self) -> None:
             index=dates,
         )
         result = fe.time_features(pd.Series(values, index=dates, name="val"))
-        assert_frame_equal(expected, result)
+        assert_frame_equal(expected, result, check_like=True, check_dtype=False)
 
     def test_timestamp_time_features(self) -> None:
         t = pd.Timestamp("2021-01-01T02:03:04.5678", tz="US/Pacific")

kats/utils/decomposition.py

@@ -188,7 +188,7 @@ def __decompose_STL(self, original: pd.DataFrame) -> Dict[str, pd.DataFrame]:
 
         # pyre-fixme[16]: Module `seasonal` has no attribute `STL`.
         result = STL(
-            data,
+            data.squeeze(),
             period=period,
             seasonal=self.seasonal,
             trend=self.trend,