def test_ice_binary(self, titanic_data, titanic_model, titanic_features):
# binary feature
grid_results, _data = _calc_ice_lines(feature_grid=0,
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature='Sex',
feature_type='binary',
predict_kwds={},
data_transformer=None,
unit_test=True)
expected = pd.DataFrame({
0: {
0: 0.527877688407898,
150: 0.8060303926467896,
300: 0.7201764583587646,
450: 0.9118255376815796,
600: 0.8612496852874756,
750: 0.9331579208374023
}
})
assert_frame_equal(grid_results.iloc[[0, 150, 300, 450, 600, 750]],
expected,
check_like=True,
check_dtype=False)
assert_array_equal(_data['Sex'].unique(), np.array([0]))
def test_ice_predict_kwds(self, titanic_data, titanic_model,
titanic_features):
# with predict_kwds
grid_results, _ = _calc_ice_lines(feature_grid=0,
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature='Sex',
feature_type='binary',
predict_kwds={'ntree_limit': 10},
data_transformer=None,
unit_test=True)
expected = pd.DataFrame({
0: {
0: 0.5039686560630798,
150: 0.6007370352745056,
300: 0.5556174516677856,
450: 0.643494725227356,
600: 0.643494725227356,
750: 0.65798020362854
}
})
assert_frame_equal(grid_results.iloc[[0, 150, 300, 450, 600, 750]],
expected,
check_like=True,
check_dtype=False)
def test_ice_numeric(self, titanic_data, titanic_model, titanic_features):
# numeric feature
grid_results, _data = _calc_ice_lines(feature_grid=10,
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature='Fare',
feature_type='numeric',
predict_kwds={},
data_transformer=None,
unit_test=True)
expected = pd.DataFrame({
10: {
0: 0.10624270886182785,
150: 0.09951823949813843,
300: 0.6190056204795837,
450: 0.16398519277572632,
600: 0.7467048764228821,
750: 0.868721067905426
}
})
assert_frame_equal(grid_results.iloc[[0, 150, 300, 450, 600, 750]],
expected,
check_like=True,
check_dtype=False)
assert_array_equal(_data['Fare'].unique(), np.array([10]))
def test_ice_onehot(self, titanic_data, titanic_model, titanic_features):
# onehot encoding feature
grid_results, _data = _calc_ice_lines(
feature_grid='Embarked_C',
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature=['Embarked_C', 'Embarked_S', 'Embarked_Q'],
feature_type='onehot',
predict_kwds={},
data_transformer=None,
unit_test=True)
expected = pd.DataFrame({
'Embarked_C': {
0: 0.19760717451572418,
150: 0.11059149354696274,
300: 0.7139607667922974,
450: 0.2575017809867859,
600: 0.9045996069908142,
750: 0.9531968832015991
}
})
assert_frame_equal(grid_results.iloc[[0, 150, 300, 450, 600, 750]],
expected,
check_like=True,
check_dtype=False)
assert_array_equal(
_data[['Embarked_C', 'Embarked_S', 'Embarked_Q']].mean().values,
np.array([1, 0, 0]))
def test_ice_data_transformer(self, titanic_data, titanic_model,
titanic_features):
# with data_transformer
def embark_change(df):
df.loc[df['Embarked_C'] == 1, 'Fare'] = 10
df.loc[df['Embarked_S'] == 1, 'Fare'] = 20
df.loc[df['Embarked_Q'] == 1, 'Fare'] = 30
return df
grid_results, _data = _calc_ice_lines(
feature_grid='Embarked_C',
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature=['Embarked_C', 'Embarked_S', 'Embarked_Q'],
feature_type='onehot',
predict_kwds={},
data_transformer=embark_change,
unit_test=True)
expected = pd.DataFrame({
'Embarked_C': {
0: 0.20869030058383942,
150: 0.10480280220508575,
300: 0.6179739832878113,
450: 0.18637187778949738,
600: 0.8106594085693359,
750: 0.8973860740661621
}
})
assert_frame_equal(grid_results.iloc[[0, 150, 300, 450, 600, 750]],
expected,
check_like=True,
check_dtype=False)
assert_array_equal(_data['Fare'].unique(), np.array([10.]))
def test_calc_ice_lines_regression(ross_data, ross_model, ross_features):
grid_results, _data = _calc_ice_lines(feature_grid=1,
data=ross_data,
model=ross_model,
model_features=ross_features,
n_classes=0,
feature='SchoolHoliday',
feature_type='binary',
predict_kwds={},
data_transformer=None,
unit_test=True)
assert_array_equal(_data['SchoolHoliday'].unique(), np.array([1]))
expected = pd.DataFrame({
1: {
0: 8802.910080560769,
100000: 8293.287914628107,
200000: 5352.321273982288,
300000: 5412.1717528683475,
400000: 7933.070072150073,
500000: 7520.956055932758,
600000: 5493.134809064146,
700000: 5528.43699339258,
800000: 4877.434213535265
}
})
assert_frame_equal(grid_results.iloc[[
0, 100000, 200000, 300000, 400000, 500000, 600000, 700000, 800000
]],
expected,
check_like=True,
check_dtype=False)
def test_calc_ice_lines_regression(ross_data, ross_model, ross_features):
grid_results, _data = _calc_ice_lines(
feature_grid=1,
data=ross_data,
model=ross_model,
model_features=ross_features,
n_classes=0,
feature="SchoolHoliday",
feature_type="binary",
predict_kwds={},
data_transformer=None,
unit_test=True,
)
assert_array_equal(_data["SchoolHoliday"].unique(), np.array([1]))
def test_ice_predict_kwds(self, titanic_data, titanic_model,
titanic_features):
# with predict_kwds
grid_results, _ = _calc_ice_lines(
feature_grid=0,
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature="Sex",
feature_type="binary",
predict_kwds={"ntree_limit": 10},
data_transformer=None,
unit_test=True,
)
def test_ice_numeric(self, titanic_data, titanic_model, titanic_features):
# numeric feature
grid_results, _data = _calc_ice_lines(
feature_grid=10,
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature="Fare",
feature_type="numeric",
predict_kwds={},
data_transformer=None,
unit_test=True,
)
assert_array_equal(_data["Fare"].unique(), np.array([10]))
def test_ice_binary(self, titanic_data, titanic_model, titanic_features):
# binary feature
grid_results, _data = _calc_ice_lines(
feature_grid=0,
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature="Sex",
feature_type="binary",
predict_kwds={},
data_transformer=None,
unit_test=True,
)
assert_array_equal(_data["Sex"].unique(), np.array([0]))
def test_calc_ice_lines_multiclass(otto_data, otto_model, otto_features):
grid_results, _data = _calc_ice_lines(
feature_grid=1,
data=otto_data,
model=otto_model,
model_features=otto_features,
n_classes=9,
feature="feat_67",
feature_type="numeric",
predict_kwds={},
data_transformer=None,
unit_test=True,
)
assert len(grid_results) == 9
assert_array_equal(_data["feat_67"].unique(), np.array([1]))
def test_ice_onehot(self, titanic_data, titanic_model, titanic_features):
# onehot encoding feature
grid_results, _data = _calc_ice_lines(
feature_grid="Embarked_C",
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature=["Embarked_C", "Embarked_S", "Embarked_Q"],
feature_type="onehot",
predict_kwds={},
data_transformer=None,
unit_test=True,
)
assert_array_equal(
_data[["Embarked_C", "Embarked_S", "Embarked_Q"]].mean().values,
np.array([1, 0, 0]),
)
def test_calc_ice_lines_multiclass(otto_data, otto_model, otto_features):
grid_results, _data = _calc_ice_lines(
feature_grid=1, data=otto_data, model=otto_model, model_features=otto_features, n_classes=9,
feature='feat_67', feature_type='numeric', predict_kwds={}, data_transformer=None, unit_test=True)
assert len(grid_results) == 9
assert_array_equal(_data['feat_67'].unique(), np.array([1]))
expected_target_0 = pd.DataFrame({1: {0: 0.56, 10000: 0.0, 20000: 0.0, 30000: 0.01, 40000: 0.01,
50000: 0.03, 60000: 0.04}})
assert_frame_equal(grid_results[0].iloc[[0, 10000, 20000, 30000, 40000, 50000, 60000]],
expected_target_0, check_like=True, check_dtype=False)
expected_target_3 = pd.DataFrame({1: {0: 0.0, 10000: 0.04, 20000: 0.03, 30000: 0.02, 40000: 0.0,
50000: 0.0, 60000: 0.0}})
assert_frame_equal(grid_results[3].iloc[[0, 10000, 20000, 30000, 40000, 50000, 60000]],
expected_target_3, check_like=True, check_dtype=False)
expected_target_7 = pd.DataFrame({1: {0: 0.0, 10000: 0.03, 20000: 0.01, 30000: 0.02, 40000: 0.03,
50000: 0.9, 60000: 0.02}})
assert_frame_equal(grid_results[7].iloc[[0, 10000, 20000, 30000, 40000, 50000, 60000]],
expected_target_7, check_like=True, check_dtype=False)
def test_ice_data_transformer(self, titanic_data, titanic_model,
titanic_features):
# with data_transformer
def embark_change(df):
df.loc[df["Embarked_C"] == 1, "Fare"] = 10
df.loc[df["Embarked_S"] == 1, "Fare"] = 20
df.loc[df["Embarked_Q"] == 1, "Fare"] = 30
return df
grid_results, _data = _calc_ice_lines(
feature_grid="Embarked_C",
data=titanic_data,
model=titanic_model,
model_features=titanic_features,
n_classes=2,
feature=["Embarked_C", "Embarked_S", "Embarked_Q"],
feature_type="onehot",
predict_kwds={},
data_transformer=embark_change,
unit_test=True,
)
assert_array_equal(_data["Fare"].unique(), np.array([10.0]))
