shap--shap
wehub-resource-sync
121 行
3.7 KiB
Python
121 行
3.7 KiB
Python
import matplotlib.pyplot as plt
|
|
import numpy as np
|
|
import pandas as pd
|
|
import pytest
|
|
from sklearn.neighbors import KNeighborsClassifier
|
|
from sklearn.tree import DecisionTreeRegressor
|
|
|
|
import shap
|
|
from shap.plots import _style
|
|
|
|
|
|
def test_waterfall_input_is_explanation():
|
|
"""Checks an error is raised if a non-Explanation object is passed as input."""
|
|
with pytest.raises(
|
|
TypeError,
|
|
match="waterfall plot requires an `Explanation` object",
|
|
):
|
|
_ = shap.plots.waterfall(np.random.randn(20, 5), show=False)
|
|
|
|
|
|
def test_waterfall_wrong_explanation_shape(explainer):
|
|
explanation = explainer(explainer.data)
|
|
|
|
emsg = "waterfall plot can currently only plot a single explanation"
|
|
with pytest.raises(ValueError, match=emsg):
|
|
shap.plots.waterfall(explanation, show=False)
|
|
|
|
|
|
@pytest.mark.mpl_image_compare(tolerance=3)
|
|
def test_waterfall(explainer):
|
|
"""Test the new waterfall plot."""
|
|
fig = plt.figure()
|
|
explanation = explainer(explainer.data)
|
|
shap.plots.waterfall(explanation[0], show=False)
|
|
plt.tight_layout()
|
|
return fig
|
|
|
|
|
|
@pytest.mark.mpl_image_compare(tolerance=3)
|
|
def test_waterfall_legacy(explainer):
|
|
"""Test the old waterfall plot."""
|
|
shap_values = explainer.shap_values(explainer.data)
|
|
fig = plt.figure()
|
|
shap.plots._waterfall.waterfall_legacy(explainer.expected_value, shap_values[0], show=False)
|
|
plt.tight_layout()
|
|
return fig
|
|
|
|
|
|
@pytest.mark.mpl_image_compare(tolerance=3)
|
|
def test_waterfall_bounds(explainer):
|
|
"""Test the waterfall plot with upper and lower error bounds plotted."""
|
|
fig = plt.figure()
|
|
explanation = explainer(explainer.data)
|
|
explanation._s.lower_bounds = explanation.values - 0.1
|
|
explanation._s.upper_bounds = explanation.values + 0.1
|
|
shap.plots.waterfall(explanation[0])
|
|
plt.tight_layout()
|
|
return fig
|
|
|
|
|
|
@pytest.mark.mpl_image_compare(tolerance=5)
|
|
def test_waterfall_custom_style(explainer):
|
|
"""Test the waterfall plot in the context of custom styles"""
|
|
|
|
# Note: the tolerance is set to 5 because matplotlib 3.10 changed the way negative values are displayed
|
|
# There is now an increased space before the negative sign, which leads to a RMS diff of ~4.4
|
|
# See: GH #3946
|
|
|
|
# TODO: reset tolerance to 3 when python 3.9 is dropped, and all tests use matplotlib 3.10+
|
|
with _style.style_context(
|
|
primary_color_positive="#9ACD32",
|
|
primary_color_negative="#FFA500",
|
|
text_color="black",
|
|
hlines_color="red",
|
|
vlines_color="red",
|
|
tick_labels_color="red",
|
|
):
|
|
fig = plt.figure()
|
|
explanation = explainer(explainer.data)
|
|
shap.plots.waterfall(explanation[0])
|
|
plt.tight_layout()
|
|
return fig
|
|
|
|
|
|
def test_waterfall_plot_for_decision_tree_explanation():
|
|
# Regression tests for GH #3129
|
|
X = pd.DataFrame({"A": [1, 2, 3], "B": [2, 1, 3]})
|
|
y = pd.Series([1, 2, 3])
|
|
model = DecisionTreeRegressor()
|
|
model.fit(X, y)
|
|
explainer = shap.TreeExplainer(model)
|
|
explanation = explainer(X)
|
|
shap.plots.waterfall(explanation[0], show=False)
|
|
|
|
|
|
def test_waterfall_plot_for_data_with_number_columns():
|
|
# GH 4150
|
|
model = KNeighborsClassifier()
|
|
|
|
def f(x):
|
|
return model.predict_proba(x)[:, 1]
|
|
|
|
X = pd.DataFrame(
|
|
[
|
|
[1, 0, 0, 0, 0],
|
|
[0, 1, 0, 0, 0],
|
|
[0, 0, 1, 0, 0],
|
|
[0, 0, 0, 1, 0],
|
|
[0, 0, 0, 0, 1],
|
|
[0, 0, 0, 1, 0],
|
|
[0, 0, 1, 0, 0],
|
|
[0, 1, 0, 0, 0],
|
|
]
|
|
)
|
|
y = pd.Series([0, 1, 2, 3, 4, 3, 2, 1])
|
|
model.fit(X, y)
|
|
med = X.median().values.reshape((1, X.shape[1]))
|
|
explainer = shap.Explainer(f, med)
|
|
shap_values = explainer(X)
|
|
shap.plots.waterfall(shap_values[0], show=False)
|