def test_transform_unseen(self, X, y, handle_unseen, expected):
enc = TargetEncoder(cols=['cat'], handle_unseen=handle_unseen)
X = pd.DataFrame(X, columns=['cat'])
enc.fit(X, pd.Series(y))
X.iloc[0, 0] = 'foo'
result = enc.transform(X)
assert_array_almost_equal(result, pd.DataFrame(expected), decimal=3)
def target_encoding(cls, X, Y=None, encoder=None):
cols = ['some_id', 'other_id']
if encoder is None:
encoder = TargetEncoder(cols=cols, min_samples=5, smoothing=5)
encoder.fit(X, Y)
encoded = encoder.transform(X).rename(
columns={c: 'tgt_enc_{}'.format(c)
for c in cols})
return pd.concat([X[cols], encoded], axis=1), encoder
def test_encode_nans(self, X, y, expected, columns):
enc = TargetEncoder(cols=['cat'])
result = enc.fit_transform(pd.DataFrame(X, columns=['cat']), pd.Series(y))
assert_array_almost_equal(result, pd.DataFrame(expected), decimal=3)
ok_('cat' in enc._mapping)
ok_(isinstance(enc._mapping['cat'], pd.DataFrame))
eq_(enc._mapping['cat'].index[0], -99999)
assert_array_equal(enc._mapping['cat'].index[1:], columns)
assert_array_equal(enc._mapping['cat'].columns, ['mean', 'count', 'value'])
def test_encode_col(self, X, y, min_samples, smoothing, expected, imputed, columns):
enc = TargetEncoder(cols=['cat'], min_samples=min_samples, smoothing=smoothing)
result = enc.fit_transform(pd.DataFrame(X, columns=['cat']), pd.Series(y))
assert_array_almost_equal(result, pd.DataFrame(expected), decimal=3)
eq_(enc._imputed, imputed)
ok_('cat' in enc._mapping)
ok_(isinstance(enc._mapping['cat'], pd.DataFrame))
assert_array_equal(enc._mapping['cat'].index, columns)
assert_array_equal(enc._mapping['cat'].columns, ['mean', 'count', 'value'])
def test_encode_multiple_cols(self, X, y, expected):
enc = TargetEncoder(cols=['cat1', 'cat2'])
result = enc.fit_transform(pd.DataFrame(X, columns=['cat1', 'cat2']), pd.Series(y))
assert_array_almost_equal(result, pd.DataFrame(expected), decimal=2)
ok_('cat1' in enc._mapping)
ok_('cat2' in enc._mapping)
ok_(isinstance(enc._mapping['cat1'], pd.DataFrame))
ok_(isinstance(enc._mapping['cat2'], pd.DataFrame))
assert_array_equal(enc._mapping['cat1'].index, ['a', 'b'])
assert_array_equal(enc._mapping['cat2'].index, ['bar', 'foo'])
assert_array_equal(enc._mapping['cat1'].columns, ['mean', 'count', 'value'])
assert_array_equal(enc._mapping['cat2'].columns, ['mean', 'count', 'value'])
def test_init(self, kwargs, cols, handle_unseen, min_samples, smoothing):
enc = TargetEncoder(**kwargs)
eq_(enc.cols, cols)
eq_(enc.handle_unseen, handle_unseen)
eq_(enc.min_samples, min_samples)
eq_(enc.smoothing, smoothing)
eq_(enc._imputed, None)
eq_(enc._mapping, {})
def test_transform_before_fit(self):
enc = TargetEncoder()
assert_raises(ValueError, enc.transform, 1)
def test_transform_error(self, X, y, expected):
enc = TargetEncoder(cols=['cat'], handle_unseen='error')
X = pd.DataFrame(X, columns=['cat'])
enc.fit(X, pd.Series(y))
X.iloc[0, 0] = 'foo'
assert_raises(ValueError, enc.transform, X)