def test_gamma_regressor(self):
model = GammaRegressor()
X = np.array([[1, 2], [2, 3], [3, 4], [4, 3]])
y = np.array([19, 26, 33, 30])
model.fit(X, y)
test_x = np.array([[1, 0], [2, 8]])
model_onnx = convert_sklearn(
model,
"scikit-learn Gamma Regressor",
[("input", FloatTensorType([None, X.shape[1]]))],
target_opset=TARGET_OPSET)
self.assertIsNotNone(model_onnx)
dump_data_and_model(test_x.astype(np.float32),
model,
model_onnx,
basename="SklearnGammaRegressor")
#
# Note:
#
# - We filter out ``ClaimAmount == 0`` as the Gamma distribution has support
# on :math:`(0, \infty)`, not :math:`[0, \infty)`.
# - We use ``ClaimNb`` as `sample_weight` to account for policies that contain
# more than one claim.
mask_train = df_train["ClaimAmount"] > 0
mask_test = df_test["ClaimAmount"] > 0
glm_sev = GammaRegressor(alpha=10.0, max_iter=10000)
glm_sev.fit(
X_train[mask_train.values],
df_train.loc[mask_train, "AvgClaimAmount"],
sample_weight=df_train.loc[mask_train, "ClaimNb"],
)
scores = score_estimator(
glm_sev,
X_train[mask_train.values],
X_test[mask_test.values],
df_train[mask_train],
df_test[mask_test],
target="AvgClaimAmount",
weights="ClaimNb",
)
print("Evaluation of GammaRegressor on target AvgClaimAmount")
print(scores)