def get_target_transformations():
transformations = []
transformations.append(
Feature(
input='total_cases',
transformer=IdentityTransformer(),
))
return transformations
def test_bad_feature_input(self):
# bad input
feature = Feature(
input=3,
transformer=sklearn.preprocessing.Imputer(),
)
validator = FeatureValidator(self.X, self.y)
result, failures = validator.validate(feature)
self.assertFalse(result)
self.assertIn('has_correct_input_type', failures)
def test_good_feature(self):
feature = Feature(
input='size',
transformer=sklearn.preprocessing.Imputer(),
)
validator = FeatureValidator(self.X, self.y)
result, failures = validator.validate(feature)
self.assertTrue(result)
self.assertEqual(len(failures), 0)
def test_bad_feature_transform_errors(self):
# transformer throws errors
feature = Feature(
input='size',
transformer=FragileTransformer(
(lambda x: True, ), (RuntimeError, ))
)
validator = FeatureValidator(self.X, self.y)
result, failures = validator.validate(feature)
self.assertFalse(result)
self.assertIn('can_transform', failures)
def test_bad_feature_deepcopy_fails(self):
class _CopyFailsTransformer(IdentityTransformer):
def __deepcopy__(self):
raise RuntimeError
feature = Feature(
input='size',
transformer=_CopyFailsTransformer(),
)
validator = FeatureValidator(self.X, self.y)
result, failures = validator.validate(feature)
self.assertFalse(result)
self.assertIn('can_deepcopy', failures)
def test_bad_feature_wrong_transform_length(self):
class _WrongLengthTransformer(
BaseEstimator, NoFitMixin, TransformerMixin):
def transform(self, X, **transform_kwargs):
new_shape = list(X.shape)
new_shape[0] += 1
output = np.arange(np.prod(new_shape)).reshape(new_shape)
return output
# doesn't return correct length
feature = Feature(
input='size',
transformer=_WrongLengthTransformer(),
)
validator = FeatureValidator(self.X, self.y)
result, failures = validator.validate(feature)
self.assertFalse(result)
self.assertIn('has_correct_output_dimensions', failures)
import sklearn.preprocessing
from fhub_core.feature import Feature
from fhub_transformers.missing import LagImputer
features = [
Feature(
input='ndvi_nw',
transformer=[
LagImputer(groupby_kwargs={'level': 'city'}),
sklearn.preprocessing.Imputer(),
sklearn.preprocessing.StandardScaler(),
],
),
]
def get_feature_transformations():
features = []
features.append(
Feature(input='ndvi_se',
transformer=[
LagImputer(groupby_kwargs={'level': 'city'}),
sklearn.preprocessing.Imputer(),
sklearn.preprocessing.StandardScaler(),
]))
features.append(
Feature(input='ndvi_sw',
transformer=[
LagImputer(groupby_kwargs={'level': 'city'}),
sklearn.preprocessing.Imputer(),
sklearn.preprocessing.StandardScaler(),
]))
features.append(
Feature(input='precipitation_amt_mm',
transformer=[
LagImputer(groupby_kwargs={'level': 'city'}),
sklearn.preprocessing.Imputer(),
SimpleFunctionTransformer(np.log1p)
]))
features.append(
Feature(input='precipitation_amt_mm', transformer=[
NullIndicator(),
]))
# Same basic cleaning of time series features.
for input_ in [
'reanalysis_air_temp_k',
'reanalysis_avg_temp_k',
'reanalysis_dew_point_temp_k',
'reanalysis_max_air_temp_k',
'reanalysis_min_air_temp_k',
'reanalysis_precip_amt_kg_per_m2',
'reanalysis_relative_humidity_percent',
'reanalysis_specific_humidity_g_per_kg',
'reanalysis_tdtr_k',
'station_avg_temp_c',
'station_diur_temp_rng_c',
'station_max_temp_c',
'station_min_temp_c',
'station_precip_mm',
]:
features.append(
Feature(input=input_,
transformer=[
LagImputer(groupby_kwargs={'level': 'city'}),
NullFiller(replacement=0.0),
sklearn.preprocessing.StandardScaler(),
]))
features.append(
Feature(input='reanalysis_sat_precip_amt_mm',
transformer=[
SingleLagger(1, groupby_kwargs={'level': 'city'}),
LagImputer(groupby_kwargs={'level': 'city'}),
NullFiller(replacement=0.0),
]))
features.append(
Feature(input=[
'reanalysis_sat_precip_amt_mm',
'reanalysis_relative_humidity_percent',
'reanalysis_specific_humidity_g_per_kg',
'reanalysis_precip_amt_kg_per_m2',
'precipitation_amt_mm',
'station_precip_mm',
],
transformer=[
LagImputer(groupby_kwargs={'level': 'city'}),
sklearn.preprocessing.Imputer(),
sklearn.decomposition.PCA(n_components=2),
]))
# add contributed features
contrib_features = get_contrib_features()
features.extend(contrib_features)
return features