def transform(self, X, **transform_args):
check_is_fitted(self, 'features_to_transform_')
if isinstance(X, pd.DataFrame):
X = X.copy()
X.loc[:, self.features_to_transform_] = self.trans(
X.loc[:, self.features_to_transform_])
return X
elif np.ndim(X) == 1:
return self.trans(X) if self.features_to_transform_ else X
elif isinstance(X, np.ndarray):
if self.features_to_transform_.any():
X = X.copy()
X = X.astype('float')
mask = np.tile(self.features_to_transform_, (X.shape[0], 1))
np.putmask(X, mask,
self.trans(X[:, self.features_to_transform_]))
return X
elif not self.features_to_transform_:
# if we wouldn't otherwise have known what to do, we can pass
# through X if transformation was not necessary anyways
return X
else:
raise TypeError(
"Couldn't apply transformer on features in {}.".format(
get_arr_desc(X)))
def transform(self, X, **transform_kwargs):
msg = "Couldn't convert object {} to named 1d DataFrame."
if isinstance(X, pd.Index):
return X.to_series().to_frame(name=self.name)
elif isinstance(X, pd.Series):
return X.to_frame(name=self.name)
elif isinstance(X, pd.DataFrame):
if X.shape[1] == 1:
X = X.copy()
X.columns = [self.name]
return X
else:
raise ValueError(msg.format(get_arr_desc(X)))
elif isinstance(X, np.ndarray):
if X.ndim == 1:
return pd.DataFrame(data=X.reshape(-1, 1), columns=[self.name])
elif X.ndim == 2 and X.shape[1] == 1:
return pd.DataFrame(data=X, columns=[self.name])
else:
raise ValueError(msg.format(get_arr_desc(X)))
raise TypeError(msg.format(get_arr_desc(X)))
def test_get_arr_desc_object():
obj = object()
expected = 'object <no shape>'
actual = get_arr_desc(obj)
assert actual == expected
def test_get_arr_desc_frame():
df = pd.DataFrame()
expected = 'DataFrame (0, 0)'
actual = get_arr_desc(df)
assert actual == expected
def test_get_arr_desc_array():
a = np.ones((2, 2))
expected = 'ndarray (2, 2)'
actual = get_arr_desc(a)
assert actual == expected