Skip to content

Commit

Permalink
reintroduce and fix firstmin_ac tests
Browse files Browse the repository at this point in the history 
Summary:
Once we migrate to LLVM17, we can fix and reintroduce tests for tsfeature firstmin_ac.

The computation for this feature was incorrect under LLVM15, which was verified by hand. See D64574622 for how to help verify

Reviewed By: akindyakov

Differential Revision: D64555619

fbshipit-source-id: 90ce3c510d96aae7ee888eb394a35f90c0e1dff2
  • Loading branch information
@islijepcevic @facebook-github-bot
islijepcevic authored and facebook-github-bot committed Oct 18, 2024
1 parent eca0d4b commit 416b4d5
Showing 1 changed file with 81 additions and 83 deletions.
164 changes: 81 additions & 83 deletions kats/tests/tsfeatures/test_tsfeatures.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -111,8 +111,6 @@ def assertDictAlmostEqual(

self.assertEqual(expected.keys(), features.keys())
for k, v in expected.items():
if k in ["firstmin_ac", "firstzero_ac"]:
continue
if isinstance(v, float):
if np.isnan(v):
self.assertTrue(np.isnan(features[k]), msg=f"{k} differ")
Expand Down Expand Up @@ -179,8 +177,8 @@ def test_tsfeatures_basic(self, test_name: str, ts_name: str) -> None:
"seas_acf1": -0.1483,
"seas_pacf1": -0.0064,
# special_ac
"firstmin_ac": 2.0,
"firstzero_ac": 4.0,
"firstmin_ac": 4,
"firstzero_ac": 4,
# holt_params
"holt_alpha": 0.0,
"holt_beta": 0.0,
Expand All @@ -202,84 +200,84 @@ def test_tsfeatures_basic(self, test_name: str, ts_name: str) -> None:
for feature_vector in feature_list:
self.assertDictAlmostEqual(expected, feature_vector)

# def test_tsfeatures(self) -> None:
# feature_vector = _univariate_features(TsFeatures().transform(self.TSData))

# feature_vector_round = {
# key: round(feature_vector[key], 6) for key in feature_vector
# }

# # test there is no nan in feature vector
# self.assertEqual(
# np.isnan(np.asarray(list(feature_vector.values()))).any(),
# False,
# )

# # test there are 40 features in the feature vector now
# self.assertEqual(
# len(np.asarray(list(feature_vector.values()))) == 40,
# True,
# )

# # test feature vector value
# rounded_truth = {
# "length": 144,
# "mean": 280.298611,
# "var": 14291.973331,
# "entropy": 0.428737,
# "lumpiness": 3041164.562906,
# "stability": 12303.627267,
# "flat_spots": 2,
# "hurst": -0.080233,
# "std1st_der": 27.206288,
# "crossing_points": 7,
# "binarize_mean": 0.444444,
# "unitroot_kpss": 0.128475,
# "heterogeneity": 126.064506,
# "histogram_mode": 155.8,
# "linearity": 0.853638,
# "trend_strength": 0.9681,
# "seasonality_strength": 0.440863,
# "spikiness": 33.502886,
# "peak": 6,
# "trough": 3,
# "level_shift_idx": 118,
# "level_shift_size": 15.6,
# "y_acf1": 0.948047,
# "y_acf5": 3.392072,
# "diff1y_acf1": 0.302855,
# "diff1y_acf5": 0.259459,
# "diff2y_acf1": -0.191006,
# "diff2y_acf5": 0.134207,
# "y_pacf5": 1.003288,
# "diff1y_pacf5": 0.219412,
# "diff2y_pacf5": 0.26101,
# "seas_acf1": 0.662904,
# "seas_pacf1": 0.15617,
# "firstmin_ac": 8,
# "firstzero_ac": 52,
# "holt_alpha": 1.0,
# "holt_beta": 0.0,
# "hw_alpha": 0.842106,
# "hw_beta": 0.052631,
# "hw_gamma": 0.157901,
# }
# if statsmodels.version >= "0.12":
# rounded_truth["trend_strength"] = 0.93833
# rounded_truth["seasonality_strength"] = 0.329934
# rounded_truth["spikiness"] = 111.697325
# feature_vector_round["holt_alpha"] = np.round(
# feature_vector_round["holt_alpha"], 1
# )
# feature_vector_round["holt_beta"] = np.round(
# feature_vector_round["holt_beta"], 1
# )
# rounded_truth["holt_alpha"] = 1.0
# rounded_truth["holt_beta"] = 0.0
# rounded_truth["hw_alpha"] = 1.0
# rounded_truth["hw_beta"] = 0.0
# rounded_truth["hw_gamma"] = 0.0
# self.assertEqual(feature_vector_round, rounded_truth)
def test_tsfeatures(self) -> None:
feature_vector = _univariate_features(TsFeatures().transform(self.TSData))

feature_vector_round = {
key: round(feature_vector[key], 6) for key in feature_vector
}

# test there is no nan in feature vector
self.assertEqual(
np.isnan(np.asarray(list(feature_vector.values()))).any(),
False,
)

# test there are 40 features in the feature vector now
self.assertEqual(
len(np.asarray(list(feature_vector.values()))) == 40,
True,
)

# test feature vector value
rounded_truth = {
"length": 144,
"mean": 280.298611,
"var": 14291.973331,
"entropy": 0.428737,
"lumpiness": 3041164.562906,
"stability": 12303.627267,
"flat_spots": 2,
"hurst": -0.080233,
"std1st_der": 27.206288,
"crossing_points": 7,
"binarize_mean": 0.444444,
"unitroot_kpss": 0.128475,
"heterogeneity": 126.064506,
"histogram_mode": 155.8,
"linearity": 0.853638,
"trend_strength": 0.9681,
"seasonality_strength": 0.440863,
"spikiness": 33.502886,
"peak": 6,
"trough": 3,
"level_shift_idx": 118,
"level_shift_size": 15.6,
"y_acf1": 0.948047,
"y_acf5": 3.392072,
"diff1y_acf1": 0.302855,
"diff1y_acf5": 0.259459,
"diff2y_acf1": -0.191006,
"diff2y_acf5": 0.134207,
"y_pacf5": 1.003288,
"diff1y_pacf5": 0.219412,
"diff2y_pacf5": 0.26101,
"seas_acf1": 0.662904,
"seas_pacf1": 0.15617,
"firstmin_ac": 8,
"firstzero_ac": 52,
"holt_alpha": 1.0,
"holt_beta": 0.0,
"hw_alpha": 0.842106,
"hw_beta": 0.052631,
"hw_gamma": 0.157901,
}
if statsmodels.version >= "0.12":
rounded_truth["trend_strength"] = 0.93833
rounded_truth["seasonality_strength"] = 0.329934
rounded_truth["spikiness"] = 111.697325
feature_vector_round["holt_alpha"] = np.round(
feature_vector_round["holt_alpha"], 1
)
feature_vector_round["holt_beta"] = np.round(
feature_vector_round["holt_beta"], 1
)
rounded_truth["holt_alpha"] = 1.0
rounded_truth["holt_beta"] = 0.0
rounded_truth["hw_alpha"] = 1.0
rounded_truth["hw_beta"] = 0.0
rounded_truth["hw_gamma"] = 0.0
self.assertEqual(feature_vector_round, rounded_truth)

# pyre-fixme[56]: Pyre was not able to infer the type of the decorator `parameter...
@parameterized.expand(
Expand Down Expand Up @@ -603,7 +601,7 @@ def test_nowcasting_error(self) -> None:
"diff2y_pacf5": 4.427552,
"seas_acf1": -0.148278,
"seas_pacf1": -0.006386,
"firstmin_ac": 2,
"firstmin_ac": 4,
"firstzero_ac": 4,
"holt_alpha": 1.014757e-09,
"holt_beta": 0.0,
Expand Down

0 comments on commit 416b4d5

Please sign in to comment.