def test_kernel_gpu_cpu_shap(dtype, n_features, n_background, model):
X_train, X_test, y_train, y_test = create_synthetic_dataset(
n_samples=n_background + 3,
n_features=n_features,
test_size=3,
noise=0.1,
dtype=dtype)
mod = model().fit(X_train, y_train)
explainer, shap_values = get_shap_values(model=mod.predict,
background_dataset=X_train,
explained_dataset=X_test,
explainer=KernelExplainer)
exp_v = explainer.expected_value
fx = mod.predict(X_test)
for test_idx in range(3):
assert (np.sum(shap_values[test_idx]) -
abs(fx[test_idx] - exp_v)) <= 1e-5
if has_shap():
import shap
explainer = shap.KernelExplainer(mod.predict, cp.asnumpy(X_train))
cpu_shap_values = explainer.shap_values(cp.asnumpy(X_test))
assert np.allclose(shap_values,
cpu_shap_values,
rtol=1e-01,
atol=1e-01)
def test_kernel_gpu_cpu_shap(dtype, nfeatures, nbackground, model):
X, y = cuml.datasets.make_regression(n_samples=nbackground + 5,
n_features=nfeatures,
noise=0.1)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=5,
random_state=42)
X_train = X_train.astype(dtype)
X_test = X_test.astype(dtype)
y_train = y_train.astype(dtype)
y_test = y_test.astype(dtype)
mod = model().fit(X_train, y_train)
cu_explainer = \
cuml.experimental.explainer.KernelExplainer(model=mod.predict,
data=X_train,
is_gpu_model=True)
cu_shap_values = cu_explainer.shap_values(X_test)
exp_v = cu_explainer.expected_value
fx = mod.predict(X_test)
for test_idx in range(5):
assert (np.sum(cu_shap_values[test_idx]) -
abs(fx[test_idx] - exp_v)) <= 1e-5
if has_shap("0.37"):
import shap
explainer = shap.KernelExplainer(mod.predict, cp.asnumpy(X_train))
shap_values = explainer.shap_values(cp.asnumpy(X_test))
# note that small variances in the l1_regression with larger
# n_features, even among runs of the same explainer can cause this
# test to be flaky, better testing strategy in process.
assert np.allclose(cu_shap_values, shap_values, rtol=1e-01, atol=1e-01)
import pytest
import treelite
import numpy as np
import cupy as cp
import cudf
from cuml.experimental.explainer.tree_shap import TreeExplainer
from cuml.common.import_utils import has_xgboost, has_shap
from cuml.common.exceptions import NotFittedError
from cuml.ensemble import RandomForestRegressor as curfr
from cuml.ensemble import RandomForestClassifier as curfc
from sklearn.datasets import make_regression, make_classification
if has_xgboost():
import xgboost as xgb
if has_shap():
import shap
@pytest.mark.parametrize('objective', [
'reg:linear', 'reg:squarederror', 'reg:squaredlogerror',
'reg:pseudohubererror'
])
@pytest.mark.skipif(not has_xgboost(), reason="need to install xgboost")
@pytest.mark.skipif(not has_shap(), reason="need to install shap")
def test_xgb_regressor(objective):
n_samples = 100
X, y = make_regression(n_samples=n_samples,
n_features=8,
n_informative=8,
n_targets=1,