Detect Overfitting or Underfitting

Problem

Given training loss, training accuracy, validation loss, and validation accuracy, determine whether the model is overfitting, underfitting, or performing well (good fit).

Solution

def detect_fitting(train_loss, train_acc, val_loss, val_acc,
                   loss_threshold=0.1, acc_threshold=0.05):
    loss_gap = val_loss - train_loss
    acc_gap = train_acc - val_acc

    if train_acc < 0.6 and val_acc < 0.6:
        return "underfitting"

    if loss_gap > loss_threshold or acc_gap > acc_threshold:
        return "overfitting"

    return "good fit"

Explanation

1. Underfitting: both training and validation performance are poor. The model is too simple.
2. Overfitting: training performance is good but validation performance is significantly worse. Large gap between training and validation metrics.
3. Good fit: both training and validation metrics are close and reasonably high.
4. The loss_threshold and acc_threshold control how much gap is tolerable.
5. Remedies: underfitting -> increase model complexity; overfitting -> add regularization, more data, or reduce complexity.

Complexity

• Time: O(1)
• Space: O(1)