def test_ColumnsSelector_empty_column():
dfX = pd.DataFrame({
"cat1": ["A", "B", "A", "D"],
"cat2": ["toto", "tata", "truc", "toto"],
"num1": [0, 1, 2, 3],
"num2": [1.1, 1.5, -2, -3.5],
"num3": [-1, 1, 25, 4],
"text1": ["aa bb", "bb bb cc", "dd aa cc", "ee"],
"text2": ["a z", "b e", "d t", "a b c"],
})
dfX2 = pd.DataFrame({
"cat1": ["D", "B"],
"cat2": ["toto", "newcat"],
"num1": [5, 6],
"num2": [0.1, -5.2],
"num3": [2, -1],
"text1": ["dd ee", "aa"],
"text2": ["t a c", "z b"],
})
for col in ([], None):
selector = ColumnsSelector(columns_to_use=col)
df_res = selector.fit_transform(dfX)
assert df_res.shape == (dfX.shape[0], 0)
assert isinstance(df_res, pd.DataFrame)
assert selector.get_feature_names() == []
df_res2 = selector.transform(dfX2)
assert df_res2.shape == (dfX2.shape[0], 0)
assert isinstance(df_res2, pd.DataFrame)
def test_ColumnsSelector_array():
mat = np.array([[0, 0, 0], [0, 1, 1], [0, 0, 1], [1, 0, 0]])
# no column
for col in (None, []):
selector = ColumnsSelector(columns_to_use=col)
mat1 = selector.fit_transform(mat)
assert mat1.shape == (mat.shape[0], 0)
assert type(mat1) == type(mat)
mat1_bis = selector.transform(mat)
assert type(mat1_bis) == type(mat)
assert mat1_bis.shape == (mat.shape[0], 0)
assert len(selector.get_feature_names()) == mat1.shape[1]
# all columns
selector = ColumnsSelector(columns_to_use="all")
mat2 = selector.fit_transform(mat)
assert mat2.shape == mat.shape
assert type(mat2) == type(mat)
assert (mat2 == mat).all()
assert mat2 is mat
mat2_bis = selector.transform(mat)
assert mat2_bis is mat
assert len(selector.get_feature_names()) == mat2.shape[1]
# 1 column
selector = ColumnsSelector(columns_to_use=[1])
mat2 = selector.fit_transform(mat)
assert mat2.shape == (mat.shape[0], 1)
assert type(mat2) == type(mat)
assert (mat[:, [1]] == mat2).all()
assert len(selector.get_feature_names()) == mat2.shape[1]
# 2 column
selector = ColumnsSelector(columns_to_use=[1, 2])
mat3 = selector.fit_transform(mat)
assert mat3.shape == (mat.shape[0], 2)
assert type(mat3) == type(mat)
assert (mat[:, [1, 2]] == mat3).all()
assert len(selector.get_feature_names()) == mat3.shape[1]
def test_ColumnsSelector_dataframe():
df = pd.DataFrame(np.array([[0, 0, 0], [0, 1, 1], [0, 0, 1], [1, 0, 0]]),
columns=["a", "b", "c"])
# no columns
for col in (None, []):
selector = ColumnsSelector(columns_to_use=col)
df1 = selector.fit_transform(df)
assert df1.shape == (df.shape[0], 0)
assert type(df1) == type(df)
df1_bis = selector.transform(df)
assert type(df1_bis) == type(df)
assert df1_bis.shape == (df.shape[0], 0)
assert len(selector.get_feature_names()) == df1.shape[1]
# all columns
selector = ColumnsSelector(columns_to_use="all")
df1 = selector.fit_transform(df)
assert df1.shape == df.shape
assert type(df1) == type(df)
assert (df1 == df).all().all()
assert df1 is df
df1_bis = selector.transform(df)
assert df1_bis is df
assert len(selector.get_feature_names()) == df1.shape[1]
# 1 columns, str
selector = ColumnsSelector(columns_to_use=["a"])
df2 = selector.fit_transform(df)
assert df2.shape == (df.shape[0], 1)
assert type(df2) == type(df)
assert (df2 == df.loc[:, ["a"]]).all().all()
assert len(selector.get_feature_names()) == df2.shape[1]
# 1 columns, int
selector = ColumnsSelector(columns_to_use=[0])
df2 = selector.fit_transform(df)
assert df2.shape == (df.shape[0], 1)
assert type(df2) == type(df)
assert (df2 == df.loc[:, ["a"]]).all().all()
assert len(selector.get_feature_names()) == df2.shape[1]
# 2 columns, str
selector = ColumnsSelector(columns_to_use=["a", "c"])
df3 = selector.fit_transform(df)
assert df3.shape == (df.shape[0], 2)
assert type(df3) == type(df)
assert (df3 == df.loc[:, ["a", "c"]]).all().all()
assert len(selector.get_feature_names()) == df3.shape[1]
# 2 columns, int
selector = ColumnsSelector(columns_to_use=["a", "c"])
df3 = selector.fit_transform(df)
assert df3.shape == (df.shape[0], 2)
assert type(df3) == type(df)
assert (df3 == df.loc[:, ["a", "c"]]).all().all()
assert len(selector.get_feature_names()) == df3.shape[1]
def test_ColumnsSelector():
dfX = pd.DataFrame({
"cat1": ["A", "B", "A", "D"],
"cat2": ["toto", "tata", "truc", "toto"],
"num1": [0, 1, 2, 3],
"num2": [1.1, 1.5, -2, -3.5],
"num3": [-1, 1, 25, 4],
"text1": ["aa bb", "bb bb cc", "dd aa cc", "ee"],
"text2": ["a z", "b e", "d t", "a b c"],
})
dfX2 = pd.DataFrame({
"cat1": ["D", "B"],
"cat2": ["toto", "newcat"],
"num1": [5, 6],
"num2": [0.1, -5.2],
"num3": [2, -1],
"text1": ["dd ee", "aa"],
"text2": ["t a c", "z b"],
})
selector = ColumnsSelector(columns_to_use=["text1", "text2"])
r1 = dfX.loc[:, ["text1", "text2"]]
r2 = dfX2.loc[:, ["text1", "text2"]]
assert (selector.fit_transform(dfX) == r1).all().all()
assert (selector.transform(dfX2) == r2).all().all()
assert selector.get_feature_names() == ["text1", "text2"]
selector = ColumnsSelector(columns_to_use=np.array(["text1", "text2"]))
r1 = dfX.loc[:, ["text1", "text2"]]
r2 = dfX2.loc[:, ["text1", "text2"]]
assert (selector.fit_transform(dfX) == r1).all().all()
assert (selector.transform(dfX2) == r2).all().all()
assert selector.get_feature_names() == ["text1", "text2"]
with pytest.raises(ValueError):
selector.transform(dfX2.loc[:, ["text2", "text1"]]
) # Error because not correct number of columns
with pytest.raises(ValueError):
selector.transform(
dfX2.loc[:, ["text3", "text1"]]) # Error because text2 not in df
with pytest.raises(ValueError):
selector.transform(dfX2.values) # Error because type changes
# This error might be ignored later
### Same thing but with 'raise_if_shape_differs=False'
selector = ColumnsSelector(columns_to_use=np.array(["text1", "text2"]),
raise_if_shape_differs=False)
r1 = dfX.loc[:, ["text1", "text2"]]
r2 = dfX2.loc[:, ["text1", "text2"]]
assert (selector.fit_transform(dfX) == r1).all().all()
assert (selector.transform(dfX2) == r2).all().all()
assert selector.get_feature_names() == ["text1", "text2"]
r3 = selector.transform(
dfX2.loc[:, ["text2", "text1"]]) # Don't raise error anymore
assert r3.shape == r2.shape
assert (r3 == r2).all(axis=None)
with pytest.raises(ValueError):
r3 = selector.transform(
dfX2.loc[:, ["text3", "text1"]]
) # Still raise an error : because text2 isn't present
with pytest.raises(ValueError):
selector.transform(dfX2.values) # Error because type changes
selector = ColumnsSelector(columns_to_use=["text1", "text2", "text3"])
with pytest.raises(ValueError): # Error because 'text3' isn't present
selector.fit(dfX)
selector = ColumnsSelector(columns_to_use=["text1", "text2"])
selector.fit(dfX)
dfX3 = dfX2.copy()
del dfX3["text1"]
with pytest.raises(
ValueError): # Error because 'text1' is no longer present
selector.transform(dfX3)
dfX3 = dfX2.copy()
dfX3.columns = ["cat1", "cat2", "num1", "num2", "num3", "textAA", "text2"]
with pytest.raises(ValueError):
selector.transform(dfX3)
selector = ColumnsSelector(columns_to_use=["^text"], regex_match=True)
r1 = dfX.loc[:, ["text1", "text2"]]
r2 = dfX2.loc[:, ["text1", "text2"]]
dfX3 = dfX.loc[:,
["text2", "cat1", "cat2", "num1", "num2", "num3", "text1"
]].copy()
assert (selector.fit_transform(dfX) == r1).all().all()
assert (selector.transform(dfX2) == r2).all().all()
assert (selector.transform(dfX3) == r1).all().all()
assert selector.get_feature_names() == ["text1", "text2"]
selector = ColumnsSelector(columns_to_use=[re.compile("^text")],
regex_match=True)
r1 = dfX.loc[:, ["text1", "text2"]]
r2 = dfX2.loc[:, ["text1", "text2"]]
dfX3 = dfX.loc[:,
["text2", "cat1", "cat2", "num1", "num2", "num3", "text1"
]].copy()
assert (selector.fit_transform(dfX) == r1).all().all()
assert (selector.transform(dfX2) == r2).all().all()
assert (selector.transform(dfX3) == r1).all().all()
assert selector.get_feature_names() == ["text1", "text2"]
selector = ColumnsSelector(columns_to_use=["^text"], regex_match=False)
r1 = dfX.loc[:, ["text1", "text2"]]
r2 = dfX2.loc[:, ["text1", "text2"]]
with pytest.raises(ValueError):
selector.fit_transform(dfX)
selector2 = ColumnsSelector(columns_to_use=[5, 6])
assert (selector2.fit_transform(dfX) == r1).all().all()
assert (selector2.transform(dfX2) == r2).all().all()
selector2b = ColumnsSelector(columns_to_use=np.array([5, 6]))
assert (selector2b.fit_transform(dfX) == r1).all().all()
assert (selector2b.transform(dfX2) == r2).all().all()
with pytest.raises(ValueError):
selector2b.transform(dfX.iloc[:, 0:-1]) # missing one column
selector3 = ColumnsSelector(columns_to_use=[10, 5])
with pytest.raises(ValueError):
selector3.fit(dfX) # Error because column 10 is not here
selector3 = ColumnsSelector(columns_to_use=[5, 6])
selector3.fit(dfX)
dfX_oneless_columns = dfX.copy()
del dfX_oneless_columns["text1"]
with pytest.raises(ValueError):
selector3.transform(dfX_oneless_columns)
selector_none = ColumnsSelector(columns_to_use="all")
assert (selector_none.fit_transform(dfX) == dfX).all().all()
antiselector = ColumnsSelector(columns_to_drop=["cat1", "cat2"])
assert (antiselector.fit_transform(dfX) ==
dfX.loc[:,
["num1", "num2", "num3", "text1", "text2"]]).all().all()
assert antiselector.get_feature_names() == [
"num1", "num2", "num3", "text1", "text2"
]
antiselector = ColumnsSelector(columns_to_drop=np.array(["cat1", "cat2"]))
assert (antiselector.fit_transform(dfX) ==
dfX.loc[:,
["num1", "num2", "num3", "text1", "text2"]]).all().all()
assert antiselector.get_feature_names() == [
"num1", "num2", "num3", "text1", "text2"
]
antiselector = ColumnsSelector(columns_to_drop=["^cat"], regex_match=True)
assert (antiselector.fit_transform(dfX) ==
dfX.loc[:,
["num1", "num2", "num3", "text1", "text2"]]).all().all()
assert antiselector.get_feature_names() == [
"num1", "num2", "num3", "text1", "text2"
]
cols = ["cat1", "cat2", "num1", "num2", "num3", "text1", "text2"]
antiselector2 = ColumnsSelector(columns_to_drop=cols)
assert antiselector2.fit_transform(dfX).shape == (4, 0) # No column
assert antiselector2.transform(dfX2).shape == (2, 0)
assert antiselector2.get_feature_names() == []
cols = [0, 1, 2, 3, 4, 5, 6]
antiselector3 = ColumnsSelector(columns_to_drop=cols)
assert antiselector3.fit_transform(dfX.values).shape == (4, 0) # No column
assert antiselector3.transform(dfX2.values).shape == (2, 0) # No column
assert antiselector3.get_feature_names() == []
cols = [0, 1, 2, 3, 4, 5, 6]
antiselector3 = ColumnsSelector(columns_to_drop=np.array(cols))
assert antiselector3.fit_transform(dfX.values).shape == (4, 0) # No column
assert antiselector3.transform(dfX2.values).shape == (2, 0) # No column
assert antiselector3.get_feature_names() == []
antiselector4 = ColumnsSelector(columns_to_drop="all")
assert antiselector4.fit_transform(dfX.values).shape == (4, 0) # No column
assert antiselector4.transform(dfX2.values).shape == (2, 0)
assert antiselector4.get_feature_names() == []
antiselector5 = ColumnsSelector(columns_to_drop="all")
assert antiselector5.fit_transform(dfX).shape == (4, 0) # No column
assert antiselector5.transform(dfX2).shape == (2, 0)
assert antiselector5.get_feature_names() == []
selector3 = ColumnsSelector(columns_to_use=["num1"])
n1 = dfX.loc[:, ["num1"]]
n2 = dfX2.loc[:, ["num1"]]
# dfX_copy = dfX.copy()
r1 = selector3.fit_transform(dfX)
r2 = selector3.transform(dfX2)
assert isinstance(r1, pd.DataFrame)
assert isinstance(r2, pd.DataFrame)
assert (r1 == n1).all().all()
assert (r2 == n2).all().all()
dfrest = dfX.loc[:, ["num1", "num2", "num3", "text1", "text2"]]
dfrest2 = dfX2.loc[:, ["num1", "num2", "num3", "text1", "text2"]]
selector4 = ColumnsSelector(columns_to_drop=["cat1", "cat2"])
assert (selector4.fit_transform(dfX) == dfrest).all().all()
assert (selector4.fit_transform(dfX2) == dfrest2).all().all()
selector5 = ColumnsSelector(columns_to_drop=[0, 1])
assert (selector5.fit_transform(dfX) == dfrest).all().all()
assert (selector5.fit_transform(dfX2) == dfrest2).all().all()
selector6 = ColumnsSelector(columns_to_use=[0, 1])
xx = np.random.randn(10, 5)
xx2 = np.random.randn(3, 5)
assert np.array_equal(selector6.fit_transform(xx), xx[:, 0:2])
assert np.array_equal(selector6.fit_transform(xx2), xx2[:, 0:2])
selector7 = ColumnsSelector(columns_to_use=["num1", "num2"])
with pytest.raises(ValueError):
selector7.fit(xx)
selector_and_antiselector = ColumnsSelector(
columns_to_use=["num1", "num2", "num3"], columns_to_drop=["num3"])
assert (selector_and_antiselector.fit_transform(dfX) ==
dfX.loc[:, ["num1", "num2"]]).all().all()
assert selector_and_antiselector.get_feature_names() == ["num1", "num2"]
selector_and_antiselector2 = ColumnsSelector(columns_to_use=["num"],
columns_to_drop=["3"],
regex_match=True)
assert (selector_and_antiselector2.fit_transform(dfX) ==
dfX.loc[:, ["num1", "num2"]]).all().all()
assert selector_and_antiselector2.get_feature_names() == ["num1", "num2"]
X = np.random.randn(20, 10)
input_features = [("COL_%d" % i) for i in range(10)]
selector = ColumnsSelector(columns_to_use=[0, 1, 5, 9])
Xsubset = selector.fit_transform(X)
assert (Xsubset == X[:, [0, 1, 5, 9]]).all()
assert selector.get_feature_names() == [0, 1, 5, 9]
assert selector.get_feature_names(input_features=input_features) == [
"COL_0", "COL_1", "COL_5", "COL_9"
]
selector_with_type = ColumnsSelector(columns_to_use="object")
r1 = dfX.loc[:, ["cat1", "cat2", "text1", "text2"]]
r2 = dfX2.loc[:, ["cat1", "cat2", "text1", "text2"]]
assert (selector_with_type.fit_transform(dfX) == r1).all().all()
assert (selector_with_type.transform(dfX2) == r2).all().all()
assert selector_with_type.get_feature_names() == [
"cat1", "cat2", "text1", "text2"
]
selector_with_type = ColumnsSelector(columns_to_drop="object")
r1 = dfX.loc[:, ["num1", "num2", "num3"]]
r2 = dfX2.loc[:, ["num1", "num2", "num3"]]
assert (selector_with_type.fit_transform(dfX) == r1).all().all()
assert (selector_with_type.transform(dfX2) == r2).all().all()
assert selector_with_type.get_feature_names() == ["num1", "num2", "num3"]
selector = ColumnsSelector(columns_to_use="object",
columns_to_drop=["text1", "text2"])
r1 = dfX.loc[:, ["cat1", "cat2"]]
r2 = dfX2.loc[:, ["cat1", "cat2"]]
assert (selector.fit_transform(dfX) == r1).all().all()
assert (selector.transform(dfX2) == r2).all().all()
assert selector.get_feature_names() == ["cat1", "cat2"]