Confusion Matrix

Build the confusion matrix for a multiclass problem.

C[i, j] counts the samples whose true label is i and whose predicted label is j. The diagonal holds the correct predictions and everything off it is a specific kind of mistake, which is why the matrix tells you far more than a single accuracy number: it says which classes the model confuses.

Labels are integers in [0, num_classes). Return an integer array of shape (num_classes, num_classes).

Watch the orientation. Rows are truth and columns are predictions, so C[2, 5] means "actually a 2, called a 5".

Function signature

def confusion_matrix(y_true, y_pred, num_classes):

Example

Input

y_true = np.array([0, 1, 1, 2])
y_pred = np.array([0, 1, 2, 2])
confusion_matrix(y_true, y_pred, 3)

Output

array([[1, 0, 0],
       [0, 1, 1],
       [0, 0, 1]])

Reasoning

Three samples are correct, so they land on the diagonal. The one mistake is a true 1 predicted as 2, which adds one to C[1, 2].

Constraints and notes

• Rows are true labels and columns are predicted labels.
• Return an integer array of shape (num_classes, num_classes).