Nonlinear SVM behaviour · Data Playground

Understand the idea

RBF kernels support nonlinear boundaries. C controls the penalty/regularisation trade-off; gamma controls the locality of the kernel. Production searches C and keeps gamma="scale".

Illustrative data

Illustrative data
Column	Values
Class A	(0, 0); (0.5, 0); (-0.5, 0); (0, 0.5); (0, -0.5)
Class B	(1.5, 0); (-1.5, 0); (0, 1.5); (0, -1.5); (1, 1); (-1, -1); (1, -1); (-1, 1)

Click the highlighted code to change the result.

example_model = SVC(kernel="rbf", C=, gamma=)
example_model.fit(example_X_prepared, example_y)
example_model.n_support_  # A: 4, B: 8

C=2, gamma=1: support A=4, B=8. The zero contour separates predicted regions. Rings identify the fitted support vectors; decision scores are not probabilities.

What the syntax does

'model__C': Controls the penalty for training margin violations in the pipeline SVM.
C: 0.5, 1, 2, 10

Supplied for this practice

These objects are prepared afresh before your editor runs.

X: Feature dataframe for the declared population, preserving row indices.
X = df[['length', 'width']]
y: Target series, aligned with X.
y = df.label
X_train / X_test / y_train / y_test: Training feature rows from the declared split. Final-test feature rows from the declared split. Training targets, aligned with X_train. Final-test targets, aligned with X_test.
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)
model: Already fitted on X_train, y_train. Fitted estimator or pipeline for the requested model family. Fitted on X_train, y_train; reuse without refitting.

Outputs you create

answer: The requested result, with the values, labels and row order described by this task.

CLASS180 · first 8 rows
`length`	`width`	`label`
236.033	-0.805568	A
581.297	0.728558	A
-1511.27	-1.00866	A
99.0248	-0.24496	A
-13.0141	-0.660765	A
681.179	0.602475	A
51.1472	0.873157	A
362.131	-0.665605	A

Your task

Search C=.5,2,10 using five stratified folds.

answer

Hints

Think: C changes the margin/error trade-off and must be chosen before final testing.

Tools: GridSearchCV, StratifiedKFold and model__C.

Approach: Evaluate the three declared C values with training-only macro-F1 folds.

Explained solution

from sklearn.model_selection import GridSearchCV,StratifiedKFold
search=GridSearchCV(model,{'model__C':[.5,2,10]},cv=StratifiedKFold(5,shuffle=True,random_state=42),scoring='f1_macro').fit(X_train,y_train)
answer=search.cv_results_['mean_test_score']

The shared folds compare regularisation choices while the pipeline keeps scaling within each fit.