def test_binsizes(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=False)
self.assertCountEqual(ale_eff.loc[:, "size"], [57, 77, 66])
def test_indexname(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=False)
self.assertEqual(ale_eff.index.name, "x5")
def test_outputshape_noCI(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=False)
self.assertEqual(ale_eff.shape, (3, 2))
self.assertCountEqual(ale_eff.columns, ["eff", "size"])
def test_effvalues(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=False)
self.assertCountEqual(
np.round(ale_eff.loc[:, "eff"], 8),
[-0.05565697, 0.02367971, 0.02044105],
)
def test_bins(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=False)
self.assertCountEqual(
ale_eff.index,
['A', 'B', 'C'],
)
def test_outputshape_withCI(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=True,
C=0.9)
self.assertEqual(ale_eff.shape, (3, 4))
self.assertCountEqual(ale_eff.columns,
["eff", "size", "lowerCI_90%", "upperCI_90%"])
def test_CIvalues(self):
ale_eff = aleplot_1D_categorical(X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
include_CI=True,
C=0.9)
# assert that the first bin do not have a CI
self.assertTrue(np.isnan(ale_eff.loc[ale_eff.index[0], "lowerCI_90%"]))
self.assertTrue(np.isnan(ale_eff.loc[ale_eff.index[0], "upperCI_90%"]))
# check the values of the CI
self.assertCountEqual(
np.round(ale_eff.loc[ale_eff.index[1]:, "lowerCI_90%"], 8),
[-0.00605249, -0.00523023],
)
self.assertCountEqual(
np.round(ale_eff.loc[ale_eff.index[1]:, "upperCI_90%"], 8),
[0.05341192, 0.04611233],
)
def test_exceptions(self):
mod_not_fit = RandomForestRegressor()
# dataset should be compatible with the model
with self.assertRaises(Exception) as mod_ex_1:
aleplot_1D_categorical(
X=self.X,
model=self.model.predict,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode_custom,
)
with self.assertRaises(Exception) as mod_ex_2:
aleplot_1D_categorical(
X=self.X,
model=self.model,
feature="x5",
predictors=self.X.columns,
encode_fun=onehot_encode_custom,
)
with self.assertRaises(Exception) as mod_ex_3:
aleplot_1D_categorical(
X=self.X,
model=self.model,
feature="x5",
predictors=self.X_cleaned.columns,
encode_fun=onehot_encode,
)
mod_ex_msg = (
"""There seems to be a problem when predicting with the model.
Please check the following:
- Your model is fitted.
- The list of predictors contains the names of all the features"""
""" used for training the model.
- The encoding function takes the raw feature and returns the"""
""" right columns encoding it, including the case of a missing category.
""")
self.assertEqual(mod_ex_1.exception.args[0], mod_ex_msg)
self.assertEqual(mod_ex_2.exception.args[0], mod_ex_msg)
self.assertEqual(mod_ex_3.exception.args[0], mod_ex_msg)