Practice: Overfitting Remedies & Tuning

# Practice: diagnose training problems from loss curves

scenarios = [
    {
        "name": "Scenario A",
        "train_losses": [2.3, 1.8, 1.2, 0.8, 0.5, 0.3, 0.2, 0.15],
        "val_losses":   [2.2, 1.7, 1.1, 0.9, 0.7, 0.6, 0.65, 0.7],
    },
    {
        "name": "Scenario B",
        "train_losses": [2.3, 2.1, 2.0, 1.9, 1.8, 1.75, 1.7, 1.65],
        "val_losses":   [2.3, 2.2, 2.1, 2.0, 1.9, 1.85, 1.8, 1.75],
    },
    {
        "name": "Scenario C",
        "train_losses": [2.3, 1.5, 0.8, 0.4, 0.2, 0.1, 0.05, 0.02],
        "val_losses":   [2.2, 1.6, 1.0, 0.9, 0.95, 1.1, 1.3, 1.6],
    },
]

def diagnose(train_losses, val_losses):
    final_train = train_losses[-1]
    final_val = val_losses[-1]
    gap = final_val - final_train
    
    if final_train > 1.0 and final_val > 1.0:
        return "UNDERFITTING: both losses are high. Increase model complexity or train longer."
    elif gap > 0.5:
        return "OVERFITTING: training loss low but validation loss high. Add dropout or more data."
    elif gap < 0.2 and final_val < 1.0:
        return "GOOD FIT: training and validation losses are close and both low."
    else:
        return "WATCH: some gap developing. Monitor closely."

print("Deep Learning Training Diagnosis:")
print()

for s in scenarios:
    print(f"{s['name']}:")
    print(f"  Final train loss: {s['train_losses'][-1]:.2f}")
    print(f"  Final val loss:   {s['val_losses'][-1]:.2f}")
    print(f"  Diagnosis: {diagnose(s['train_losses'], s['val_losses'])}")
    print()