Monitoring model performance in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Monitoring model performance

Which metric matters for monitoring model performance and WHY

When we watch how a model works over time, we want to see if it keeps making good predictions. The main metrics to check are accuracy, precision, recall, and F1 score. These tell us if the model is still correct and balanced in its guesses. We also look at loss to see if the model’s errors are growing. Monitoring these helps catch problems early, like if the model starts to forget or if the data changes.

Confusion matrix example for monitoring

      Actual \ Predicted | Positive | Negative
      -------------------|----------|---------
      Positive           |    80    |   20    
      Negative           |    10    |   90    

      Total samples = 200

      Precision = 80 / (80 + 10) = 0.89
      Recall = 80 / (80 + 20) = 0.80
      Accuracy = (80 + 90) / 200 = 0.85
    

This matrix helps us see if the model is confusing positive and negative cases. Watching it over time shows if mistakes increase.

Precision vs Recall tradeoff with examples

Sometimes, improving one metric lowers the other. For example:

• Spam filter: High precision means fewer good emails marked as spam. We want to avoid losing important emails.
• Cancer detection: High recall means catching most cancer cases. Missing a case is very bad.

When monitoring, if precision drops but recall stays high, the model may mark too many good cases as bad. If recall drops, it may miss important cases. We watch these to keep the balance right.

What good vs bad metric values look like when monitoring

Good: Metrics stay stable or improve over time. For example, accuracy around 90%, precision and recall both above 85%, and loss stays low or decreases.

Bad: Metrics suddenly drop or slowly decline. For example, accuracy falls below 70%, precision or recall drop below 50%, or loss increases. This means the model may be confused or data has changed.

Common pitfalls when monitoring model performance

• Accuracy paradox: High accuracy can be misleading if data is unbalanced. For example, 95% accuracy on 95% negative data means the model ignores positives.
• Data leakage: If future data leaks into training, metrics look great but fail in real use.
• Overfitting indicators: Training metrics improve but test or live metrics worsen, showing the model learned noise, not real patterns.
• Ignoring metric trends: Small drops over time can signal problems before big failures.

Self-check question

Your model has 98% accuracy but only 12% recall on fraud cases. Is it good for production? Why or why not?

Answer: No, it is not good. Even though accuracy is high, the model misses 88% of fraud cases (low recall). This means many frauds go undetected, which is risky. For fraud detection, high recall is critical.

Key Result

Monitoring accuracy, precision, recall, and loss over time helps detect when a model's performance degrades or data changes.

Practice

1. Why is it important to monitor a machine learning model after deployment?

easy

A. To ensure the model continues to perform well on new data

B. To increase the model's training speed

C. To reduce the size of the model file

D. To change the model's algorithm automatically

Solution

1. Step 1: Understand model deployment context

   After deployment, models face new data that may differ from training data.

2. Step 2: Importance of monitoring

   Monitoring helps detect if the model's accuracy or other metrics drop over time.

3. Final Answer:

   To ensure the model continues to perform well on new data -> Option A

4. Quick Check:

   Monitoring = Ensure performance [OK]

Hint: Monitoring keeps models accurate after deployment [OK]

Common Mistakes:

• Confusing monitoring with training speed
• Thinking monitoring changes model algorithm automatically
• Assuming monitoring reduces model size

2. Which of the following is the correct way to calculate accuracy in Python given true labels y_true and predicted labels y_pred?

easy

A. accuracy = sum(y_true != y_pred) / len(y_true)

B. accuracy = len(y_true) / sum(y_true == y_pred)

C. accuracy = sum(y_true == y_pred) / len(y_true)

D. accuracy = sum(y_pred) / sum(y_true)

Solution

1. Step 1: Understand accuracy formula

   Accuracy = number of correct predictions divided by total predictions.

2. Step 2: Match formula to code

   sum(y_true == y_pred) counts correct predictions; dividing by len(y_true) gives accuracy.

3. Final Answer:

   accuracy = sum(y_true == y_pred) / len(y_true) -> Option C

4. Quick Check:

   Correct predictions / total = accuracy [OK]

Hint: Accuracy = correct predictions divided by total [OK]

Common Mistakes:

• Dividing total by correct instead of correct by total
• Using inequality instead of equality for correct count
• Summing predictions instead of comparing labels

3. Given the following Python code snippet monitoring model accuracy over time, what will be the printed output?

accuracies = [0.9, 0.85, 0.8, 0.75]
threshold = 0.8
alerts = [acc < threshold for acc in accuracies]
print(alerts)

medium

A. [False, False, True, True]

B. [True, True, True, True]

C. [True, False, False, False]

D. [False, False, False, True]

Solution

1. Step 1: Evaluate each accuracy against threshold

   Check if each accuracy is less than 0.8: 0.9 < 0.8? False; 0.85 < 0.8? False; 0.8 < 0.8? False; 0.75 < 0.8? True.

2. Step 2: Create alerts list

   List comprehension results in [False, False, False, True].

3. Final Answer:

   [False, False, False, True] -> Option D

4. Quick Check:

   Compare each value to threshold [OK]

Hint: Check each value against threshold in list comprehension [OK]

Common Mistakes:

• Misunderstanding < vs <= in comparisons
• Confusing True/False order in list
• Assuming 0.8 < 0.8 is True

4. You have this Python code to monitor model loss, but it raises an error. What is the error and how to fix it?

losses = [0.3, 0.25, 0.4, 0.35]
for i in range(len(losses))
    if losses[i] > 0.3:
        print(f"High loss at index {i}: {losses[i]}")

medium

A. SyntaxError due to missing colon after for loop; add ':' after 'range(len(losses))' to fix

B. TypeError because losses is not iterable; convert losses to list to fix

C. IndentationError due to wrong indentation; fix indentation of print statement

D. NameError because variable 'i' is undefined; define 'i' before loop

Solution

1. Step 1: Identify syntax error in for loop

   The for loop line is missing a colon at the end, which is required in Python syntax.

2. Step 2: Fix syntax by adding colon

   Add ':' after range(len(losses)) to correct the syntax and allow the loop to run.

3. Final Answer:

   SyntaxError due to missing colon after for loop; add ':' after 'range(len(losses))' to fix -> Option A

4. Quick Check:

   Missing colon = SyntaxError [OK]

Hint: Check colons after for/if statements [OK]

Common Mistakes:

• Ignoring missing colon causing syntax error
• Confusing indentation error with syntax error
• Assuming variable 'i' is undefined

5. You want to automate alerts when your model's F1 score drops below 0.7. Which approach best fits this monitoring task?

hard

A. Ignore F1 score and monitor only training loss during training

B. Set up a scheduled job that calculates F1 score daily and sends an alert if below 0.7

C. Manually check F1 score once a month and update the model if needed

D. Retrain the model automatically every hour regardless of performance

Solution

1. Step 1: Understand monitoring automation

   Automated scheduled jobs can regularly check model metrics like F1 score.

2. Step 2: Choose alerting based on threshold

   Sending alerts when F1 score drops below 0.7 helps catch performance issues early.

3. Final Answer:

   Set up a scheduled job that calculates F1 score daily and sends an alert if below 0.7 -> Option B

4. Quick Check:

   Automate checks + alerts = Set up a scheduled job that calculates F1 score daily and sends an alert if below 0.7 [OK]

Hint: Automate metric checks and alert on threshold breach [OK]

Common Mistakes:

• Retraining without checking performance first
• Manual checks too infrequent for timely alerts
• Ignoring key metrics like F1 score