def test_ale_plot():
"""Test that proper errors are raised."""
with pytest.raises(ValueError, match=r".*'model'.*'predictor'.*"):
ale_plot(model=None, train_set=pd.DataFrame([1]), features=[0])
with pytest.raises(ValueError, match=r"'3' 'features'.*"):
ale_plot(model=SimpleModel(),
train_set=pd.DataFrame([1]),
features=list(range(3)))
with pytest.raises(ValueError, match=r"'0' 'features'.*"):
ale_plot(model=SimpleModel(), train_set=pd.DataFrame([1]), features=[])
with pytest.raises(NotImplementedError,
match="'features_classes' is not implemented yet."):
ale_plot(
model=SimpleModel(),
train_set=pd.DataFrame([1]),
features=[0],
features_classes=["a"],
)
with pytest.raises(ValueError,
match=r"1 feature.*but 'bins' was not an integer."):
ale_plot(
model=SimpleModel(),
train_set=pd.DataFrame([1]),
features=[0],
bins=1.0,
)
def test_model(features, columns):
"""Given a model with a predict method, a plot should be created."""
plt.ion() # Prevent plt.show() from blocking.
np.random.seed(1)
train_set = pd.DataFrame(np.random.random((100, len(columns))),
columns=columns)
with assert_n_created_figures():
ale_plot(SimpleModel(), train_set, features)
# Clean up the created figure.
plt.close()
def test_monte_carlo(features, columns):
plt.ion() # Prevent plt.show() from blocking.
np.random.seed(1)
train_set = pd.DataFrame(np.random.random((100, len(columns))),
columns=columns)
with assert_n_created_figures():
ale_plot(
model=None,
train_set=train_set,
features=features,
predictor=simple_predictor,
monte_carlo=True,
)
# Clean up the created figure.
plt.close()
def test_predictor(features, columns):
"""Given a predictor function, a plot should be created."""
plt.ion() # Prevent plt.show() from blocking.
np.random.seed(1)
train_set = pd.DataFrame(np.random.random((100, len(columns))),
columns=columns)
with assert_n_created_figures():
ale_plot(
model=None,
train_set=train_set,
features=features,
predictor=simple_predictor,
)
# Clean up the created figure.
plt.close()
def test_df_column_features():
"""Test the handling of the `features` argument.
No matter the type of the `features` iterable, `ale_plot` should be able to select
the right columns.
"""
plt.ion() # Prevent plt.show() from blocking.
n_col = 3
np.random.seed(1)
train_set = pd.DataFrame(np.random.random((100, n_col)),
columns=list(ascii_lowercase[:n_col]))
with assert_n_created_figures():
ale_plot(SimpleModel(), train_set, train_set.columns[:1])
# Clean up the created figure.
plt.close()
shap.summary_plot(shap_values, X_train)
shap.summary_plot(shap_values, X_train, plot_type="bar")
#Applying LIME for explainability (But we created a new file.py because apparently Lime works better with lightgbm than xgboost)
#Applying ALE for explainability
from alepython import ale_plot
matplotlib.rc("figure", figsize=(9, 6))
ale_plot(
classifier,
X_test,
X_test.columns[:1],
bins=20,
monte_carlo=True,
monte_carlo_rep=100,
monte_carlo_ratio=0.6,
)
matplotlib.rc("figure", figsize=(9, 6))
ale_plot(classifier, X_test, X_test.columns[1:3], bins=10)
#Applying PDP for explainability for 1 feature
#Binary feature='Bump'
## 1.1 target distribution through feature 'Bump'
fig, axes, summary_df = info_plots.target_plot(df=train_df,
feature='Bump',
feature_name='Bump',