Problems

1 / 1

Compute Confusion Matrix with Normalization

Metrics & Evaluation

Hard

Build a $K\times K$ confusion matrix whose row is the true class and whose column is the predicted class:

C_{ij}=\sum_{n=1}^{N}\mathbf{1}[y_n=i\ \text{and}\ \hat{y}_n=j]

Here, $N$ is the sample count, $K$ is the number of classes, $y_n$ is the true label, and $\hat{y}_n$ is the predicted label. The normalize mode is "none" for counts, "true" for rows summing to one, "pred" for columns summing to one, or "all" for the entire matrix summing to one. A zero row or column remains zero. Return the matrix as a NumPy array.

Loading visualization...

Examples

Input: y_true = [0, 1, 1], y_pred = [0, 1, 0], num_classes = None, normalize = "none"

Output: [[1, 0], [1, 1]]

Explanation: One class-0 sample is correct, while the two class-1 samples split between predictions 0 and 1.

Input: y_true = [0, 1, 1], y_pred = [0, 1, 0], num_classes = None, normalize = "true"

Output: [[1, 0], [0.5, 0.5]]

Hint 1

Flatten each pair into y_true * K + y_pred before calling np.bincount.

Hint 2

Use np.divide(matrix, totals, out=..., where=totals != 0) for safe normalization.

Requirements

Count samples at their true-label row and predicted-label column
Infer the class count when num_classes is None
Apply the requested normalization without changing zero rows or columns
Return an integer NumPy array for "none" and a float NumPy array otherwise

Constraints

Labels are integers from 0 through K - 1
y_true and y_pred have equal lengths
Use NumPy only

Try Similar Problems

Classification Metrics Metrics F1 Micro Cohens Kappa Log Loss Per Sample R2 Score

Prev Next

Compute Confusion Matrix with Normalization

Metrics & Evaluation

Hard

Build a $K\times K$ confusion matrix whose row is the true class and whose column is the predicted class:

C_{ij}=\sum_{n=1}^{N}\mathbf{1}[y_n=i\ \text{and}\ \hat{y}_n=j]

Loading visualization...

Examples

Input: y_true = [0, 1, 1], y_pred = [0, 1, 0], num_classes = None, normalize = "none"

Output: [[1, 0], [1, 1]]

Explanation: One class-0 sample is correct, while the two class-1 samples split between predictions 0 and 1.

Input: y_true = [0, 1, 1], y_pred = [0, 1, 0], num_classes = None, normalize = "true"

Output: [[1, 0], [0.5, 0.5]]

Hint 1

Flatten each pair into y_true * K + y_pred before calling np.bincount.

Hint 2

Use np.divide(matrix, totals, out=..., where=totals != 0) for safe normalization.

Requirements

Count samples at their true-label row and predicted-label column
Infer the class count when num_classes is None
Apply the requested normalization without changing zero rows or columns
Return an integer NumPy array for "none" and a float NumPy array otherwise

Constraints

Labels are integers from 0 through K - 1
y_true and y_pred have equal lengths
Use NumPy only

Try Similar Problems

Classification Metrics Metrics F1 Micro Cohens Kappa Log Loss Per Sample R2 Score

Case 1

Case 2

y_true =

Accepts: array

y_pred =

Accepts: array

num_classes =

Accepts: any

normalize =

Accepts: string