def test_fit(self):
# training data
d = {"col1": ["a", "a", "c"], "col2": ["w", "e", "d"]}
df = pd.DataFrame(data=d)
lb = LabelBinarizer()
# check first column
lb.fit(df, "col1")
data_json = lb.to_json()
self.assertTrue("new_columns" in data_json)
# we take alphabetical order
self.assertTrue("col1_c" in data_json["new_columns"])
self.assertTrue("col1_a" not in data_json["new_columns"])
self.assertTrue("unique_values" in data_json)
self.assertTrue("a" in data_json["unique_values"])
self.assertTrue("c" in data_json["unique_values"])
lb = LabelBinarizer()
# check second column
lb.fit(df, "col2")
data_json = lb.to_json()
self.assertTrue("new_columns" in data_json)
self.assertTrue("col2_w" in data_json["new_columns"])
self.assertTrue("col2_e" in data_json["new_columns"])
self.assertTrue("col2_d" in data_json["new_columns"])
self.assertTrue("unique_values" in data_json)
self.assertTrue("w" in data_json["unique_values"])
self.assertTrue("e" in data_json["unique_values"])
self.assertTrue("d" in data_json["unique_values"])
def test_to_and_from_json_booleans(self):
# training data
d = {"col1": ["a", "a", "c"], "col2": [True, True, False]}
df = pd.DataFrame(data=d)
# fit binarizer
lb1 = LabelBinarizer()
lb1.fit(df, "col1")
lb2 = LabelBinarizer()
lb2.fit(df, "col2")
# test data
d_test = {
"col1": ["c", "c", "a"],
"col2": [False, False, True],
"col3": [2, 3, 4],
}
df_test = pd.DataFrame(data=d_test)
# to json and from json
new_lb1 = LabelBinarizer()
new_lb2 = LabelBinarizer()
new_lb1.from_json(lb1.to_json())
new_lb2.from_json(json.loads(json.dumps(lb2.to_json(), indent=4)))
# transform
df_test = new_lb1.transform(df_test, "col1")
df_test = new_lb2.transform(df_test, "col2")
# for binary column, only one value is left, old column should be deleted
self.assertTrue("col1_c" in df_test.columns)
self.assertTrue("col1" not in df_test.columns)
self.assertEqual(2, np.sum(df_test["col1_c"]))
# for multiple value colum, all columns should be added
self.assertTrue("col2_True" in df_test.columns)
self.assertTrue("col2" not in df_test.columns)
self.assertEqual(1, np.sum(df_test["col2_True"]))
# do not touch continuous attribute
self.assertTrue("col3" in df_test.columns)
def _fit_categorical_convert(self, X):
for column in self._columns:
if PreprocessingUtils.get_type(
X[column]) != PreprocessingUtils.CATEGORICAL:
# no need to convert, already a number
continue
# limit categories - it is needed when doing one hot encoding
# this code is also used in predict.py file
# and transform_utils.py
# TODO it needs refactoring !!!
too_much_categories = len(np.unique(list(X[column].values))) > 200
lbl = None
if (self._convert_method
== PreprocessingCategorical.CONVERT_ONE_HOT
and not too_much_categories):
lbl = LabelBinarizer()
lbl.fit(X, column)
else:
lbl = LabelEncoder()
lbl.fit(X[column])
if lbl is not None:
self._convert_params[column] = lbl.to_json()
