def test_easy_preprocessor_cat_cols():
# Create dataframe with 1 numerical and 3 categorical features.
cat_1_unique = ['a', 'b', 'c']
cat_2_unique = ['D', 'E', 'F', 'G']
cat_3_unique = [-2.0, -1.0]
data = pd.DataFrame({
'cat_1': ['a', 'b', 'a', '', '', 'c', 'a', 'c', 'a', 'a'],
'cat_2': ['D', 'D', 'E', np.NaN, 'D', np.NaN, 'E', 'F', 'F', 'G'],
'cat_3': [-2., -1., -1., -2., -1., '', -1., -2., '', np.NaN],
'num':
np.sin(range(10)), # Valid continuous feature.
})
# Preprocess data, i.e. replace empty strings with NaNs, impute NaNs, and
# encode categorical variables with OneHotEncoder.
ep = EasyPreprocessor()
data_t = ep.fit_transform(data)
cat_all = [cat_1_unique, cat_2_unique, cat_3_unique]
n_unique_cats = sum(len(cats) for cats in cat_all)
# The number of features after preprocessing must be equal to
# the number of unique categories within the dataframe + the number of
# remaining valid features.
assert data_t.shape[1] == n_unique_cats + 1
cat_pipe = ep.ct_.named_transformers_['categorical']
ohe = cat_pipe.named_steps['onehotencoder']
# The category sets detected by OneHotEncoder inside EasyPreprocessor
# must match the ones specified in cat_all.
assert len(ohe.categories_) == len(cat_all)
for cat_list, ohe_cat_list in zip(cat_all, ohe.categories_):
assert set(cat_list) == set(ohe_cat_list)
def test_simple_preprocessor_imputed_features():
# Issue: 211
data = pd.DataFrame({'A': [0, 1, 2, 1, np.NaN]}, dtype=int)
types = detect_types(data, type_hints={'A': 'categorical'})
ep = EasyPreprocessor(types=types)
ep.fit(data)
expected_names = ['A_0', 'A_1', 'A_2', 'A_imputed_False', 'A_imputed_True']
assert ep.get_feature_names() == expected_names
def test_simple_preprocessor():
sp = EasyPreprocessor()
sp.fit(X_cat)
trans = sp.transform(X_cat)
assert trans.shape == (3, 7) # FIXME should be 6?
iris = load_iris()
sp = EasyPreprocessor()
sp.fit(iris.data)
def test_boolean_and_nan(null_object):
X = pd.DataFrame({'a': [True, False, True, False, null_object]})
types = detect_types(X)
assert types.categorical.a
X_preprocessed = EasyPreprocessor().fit_transform(X)
assert X_preprocessed.shape[1] == 4
assert all(np.unique(X_preprocessed) == [0, 1])
def test_titanic_feature_names():
path = os.path.dirname(__file__)
titanic = pd.read_csv(os.path.join(path, '../datasets/titanic.csv'))
ep = EasyPreprocessor()
ep.fit(clean(titanic.drop('survived', axis=1)))
expected_names = [
'age_dabl_continuous', 'body_dabl_continuous', 'fare_dabl_continuous',
'age_?_0.0', 'age_?_1.0', 'body_?_0.0', 'body_?_1.0', 'pclass_1',
'pclass_2', 'pclass_3', 'sex_female', 'sex_male', 'embarked_?',
'embarked_C', 'embarked_Q', 'embarked_S', 'boat_1', 'boat_10',
'boat_11', 'boat_12', 'boat_13', 'boat_13 15', 'boat_13 15 B',
'boat_14', 'boat_15', 'boat_15 16', 'boat_16', 'boat_2', 'boat_3',
'boat_4', 'boat_5', 'boat_5 7', 'boat_5 9', 'boat_6', 'boat_7',
'boat_8', 'boat_8 10', 'boat_9', 'boat_?', 'boat_A', 'boat_B',
'boat_C', 'boat_C D', 'boat_D'
]
assert ep.get_feature_names() == expected_names
def test_titanic_feature_names():
path = os.path.dirname(__file__)
titanic = pd.read_csv(os.path.join(path, '../datasets/titanic.csv'))
ep = EasyPreprocessor()
ep.fit(clean(titanic.drop('survived', axis=1)))
expected_names = [
'sibsp', 'parch', 'age_dabl_continuous', 'fare_dabl_continuous',
'body_dabl_continuous', 'pclass_1', 'pclass_2', 'pclass_3',
'sex_female', 'sex_male', 'sibsp_0', 'sibsp_1', 'sibsp_2', 'sibsp_3',
'sibsp_4', 'sibsp_5', 'sibsp_8', 'parch_0', 'parch_1', 'parch_2',
'parch_3', 'parch_4', 'parch_5', 'parch_6', 'parch_9', 'embarked_?',
'embarked_C', 'embarked_Q', 'embarked_S', 'boat_1', 'boat_10',
'boat_11', 'boat_12', 'boat_13', 'boat_13 15', 'boat_13 15 B',
'boat_14', 'boat_15', 'boat_15 16', 'boat_16', 'boat_2', 'boat_3',
'boat_4', 'boat_5', 'boat_5 7', 'boat_5 9', 'boat_6', 'boat_7',
'boat_8', 'boat_8 10', 'boat_9', 'boat_?', 'boat_A', 'boat_B',
'boat_C', 'boat_C D', 'boat_D', 'age_?_0.0', 'age_?_1.0', 'body_?_0.0',
'body_?_1.0'
]
try:
assert ep.get_feature_names() == expected_names
except AssertionError:
# OHE uses int in newer versions
expected_names[57] = 'age_?_0'
expected_names[58] = 'age_?_1'
expected_names[59] = 'body_?_0'
expected_names[60] = 'body_?_1'
assert ep.get_feature_names() == expected_names
# without clean
X = ep.fit_transform(titanic.drop('survived', axis=1))
# FIXME can't do that yet
# assert ep.get_feature_names() == expected_names_no_clean
assert not np.isnan(X).any()
def test_simple_preprocessor():
sp = EasyPreprocessor()
trans = sp.fit_transform(X_cat)
assert trans.shape == (3, 6)
iris = load_iris()
sp = EasyPreprocessor()
sp.fit(iris.data)
def test_easy_preprocessor_transform():
titanic = load_titanic()
titanic_clean = clean(titanic)
X, y = titanic_clean.drop("survived", axis=1), titanic_clean.survived
X_train, X_val, y_train, y_val = train_test_split(X, y, stratify=y,
random_state=42)
pipe = make_pipeline(EasyPreprocessor(), LogisticRegression(C=0.1))
pipe.fit(X_train, y_train)
pipe.predict(X_train)
pipe.predict(X_val)
def test_simple_preprocessor_dirty_float():
dirty = pd.DataFrame(make_dirty_float())
fp = EasyPreprocessor()
fp.fit(dirty)
res = fp.transform(dirty)
assert res.shape == (100, 3)
rowsum = res.sum(axis=0)
# count of "garbage"
assert rowsum[1] == 1
# count of "missing"
assert rowsum[2] == 9
# make sure we can transform a clean column
fp.transform(pd.DataFrame(['0', '1', '2'], columns=['a_column']))
