diff --git a/examples/svm/svm.py b/examples/svm/svm.py
index 68129bb..22a90cd 100644
--- a/examples/svm/svm.py
+++ b/examples/svm/svm.py
@@ -60,7 +60,7 @@ def main(params):
     """
     # Extract the kernel name and parameters. This is just a short expression
     # to get the only dictionary entry, which should have our hyperparameters.
-    kernel = params[list(params.keys())[0]]
+    kernel_params = list(params.values())[0]
 
     # Load the training and test sets.
     (X_train, y_train), (X_test, y_test) = load_data()
@@ -68,11 +68,20 @@ def main(params):
     X_train = X_train.astype(np.float32) / 255.0
     X_test = X_test.astype(np.float32) / 255.0
 
+    # Set up the SVM with its parameterized kernel. The long form of instantiation
+    # is done here to show what `kernel_params` looks like internally.
+    # This can be shortened to `svc = SVC(**kernel_params)`
+    svc = SVC(kernel=kernel_params['kernel'],
+              C=kernel_params['C'],
+              gamma=kernel_params['gamma'] if 'gamma' in kernel_params else None,
+              coef0=kernel_params['coef0'] if 'coef0' in kernel_params else None,
+              degree=kernel_params['degree'] if 'degree' in kernel_params else None)
+
     # Set up parallel training across as many cores as are available on the
     # worker.
     s = OneVsRestClassifier(
         BaggingClassifier(
-            SVC(**kernel),
+            svc,
             n_estimators=10,
             max_samples=0.1,
             n_jobs=-1))