def test_with_feature_attachment():
X, y = make_classification_df(n_num_features=5, class_sep=0.7)
params = {
'objective': 'binary',
'max_depth': 8
}
with get_temp_directory() as temp_feature_path:
cols = list(X.columns)
for i, c in enumerate(cols):
if X.shape[1] == 1:
break
save_feature(X[[c]], i, directory=temp_feature_path)
X.drop(c, axis=1, inplace=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, shuffle=False)
with get_temp_directory() as temp_path:
result_wo_feature = run_experiment(params, X_train, y_train, X_test, logging_directory=temp_path)
with get_temp_directory() as temp_path:
result_w_feature = run_experiment(params, X_train, y_train, X_test, logging_directory=temp_path,
feature_list=[0, 1, 2, 3], feature_directory=temp_feature_path)
assert result_w_feature.metrics[-1] > result_wo_feature.metrics[-1]
def test_feature_exists():
df = pd.DataFrame({
'a': [1, 2, 3, 4, 5] + [None] * 5
})
with get_temp_directory() as tmp:
fs.save_feature(df[['a']], 0, directory=tmp)
with pytest.raises(RuntimeError):
fs.save_feature(df, 0, overwrite=False, directory=tmp)
def test_save_feature():
df = pd.DataFrame()
df['a'] = np.arange(100)
with get_temp_directory() as tmp:
fs.save_feature(df, 0, tmp)
assert os.path.exists(os.path.join(tmp, '0.f'))
def test_load_feature():
df = pd.DataFrame()
df['a'] = np.arange(100)
with get_temp_directory() as tmp:
fs.save_feature(df, 0, tmp)
df_loaded = fs.load_feature(0, tmp)
assert_frame_equal(df, df_loaded)
def test_load_features():
df = pd.DataFrame()
df['a'] = np.arange(100).astype(float)
df['b'] = np.arange(100).astype(int)
df['c'] = np.arange(100).astype(int)
with get_temp_directory() as tmp:
fs.save_feature(df[['b']], 0, tmp)
fs.save_feature(df[['c']], 1, tmp)
df_loaded = fs.load_features(df[['a']], [0, 1], tmp)
assert_frame_equal(df, df_loaded)
def test_load_feature_ignore_all_columns():
df = pd.DataFrame()
df['a'] = np.arange(100).astype(float)
df['b'] = np.arange(100).astype(int)
df['c'] = np.arange(100).astype(int)
with get_temp_directory() as tmp:
fs.save_feature(df, 0, tmp)
df_loaded = fs.load_feature(0, tmp, ignore_columns=['a', 'b', 'c', 'X'])
assert_frame_equal(df_loaded, df.drop(['a', 'b', 'c'], axis=1))
def test_various_dtypes():
df = pd.DataFrame()
df['a'] = np.arange(100).astype(float)
df['b'] = np.arange(100).astype(int)
df['c'] = np.arange(100).astype(np.uint8)
df['d'] = np.arange(100).astype(np.uint16)
df['e'] = np.arange(100).astype(np.uint32)
df['f'] = np.arange(100).astype(np.int8)
df['g'] = np.arange(100).astype(np.int16)
df['h'] = np.arange(100).astype(np.int32)
df['i'] = np.arange(100).astype(np.int64)
with get_temp_directory() as tmp:
fs.save_feature(df, 0, tmp)
df_loaded = fs.load_feature(0, tmp)
assert_frame_equal(df, df_loaded)
def test_load_features_no_base():
df = pd.DataFrame()
df['a'] = np.arange(100).astype(float)
df['b'] = np.arange(100).astype(int)
df['c'] = np.arange(100).astype(int)
with get_temp_directory() as tmp:
fs.save_feature(df[['b']], 0, tmp)
fs.save_feature(df[['c']], 1, tmp)
fs.save_feature(df[['a']], '2', tmp)
df_loaded = fs.load_features(None, [0, 1, '2'], tmp)
assert list(df_loaded.columns) == ['b', 'c', 'a']
def test_invalid_feature():
df = pd.DataFrame({
'a': [1, 2, 3, 4, 5] + [None] * 5,
'b': np.random.randint(0, 10, size=10)
})
y = pd.Series([1, 0, 1, 0, 1])
with get_temp_directory() as tmp:
with pytest.raises(RuntimeError):
fs.save_feature(df[['a']], 0, reference_target_variable=y, directory=tmp)
with pytest.raises(RuntimeError):
fs.save_feature(df, 0, reference_target_variable=y, directory=tmp)
# ok
fs.save_feature(df[['b']], 0, reference_target_variable=y, directory=tmp)
def test_load_features_duplicate_col_name():
df = pd.DataFrame()
df['a'] = np.arange(100).astype(float)
df['b'] = np.arange(100).astype(int)
df['c'] = np.arange(100).astype(int)
with get_temp_directory() as tmp:
fs.save_feature(df[['a', 'b']], 0, tmp)
fs.save_feature(df[['b', 'c']], 1, tmp)
fs.save_feature(df[['b', 'a']], 'X', tmp)
df_loaded = fs.load_features(None, [0, 1, 'X'], tmp, rename_duplicate=True)
assert list(df_loaded.columns) == ['a', 'b', 'b_1', 'c', 'b_X', 'a_X']
df_loaded = fs.load_features(None, [0, 1, 'X'], tmp, rename_duplicate=False)
assert list(df_loaded.columns) == ['a', 'b', 'b', 'c', 'b', 'a']
