Error handling and rate limits in Prompt Engineering / GenAI - Model Metrics & Evaluation

Metrics & Evaluation - Error handling and rate limits

Which metric matters for Error handling and rate limits and WHY

Error handling and rate limits focus on system reliability and user experience rather than traditional ML accuracy metrics. Key metrics include error rate (how often requests fail), latency (response time), and throughput (requests handled per second). Monitoring these helps ensure the system responds well under load and recovers gracefully from errors.

Confusion matrix or equivalent visualization

Request Outcome Confusion Table:

| Outcome       | Count |
|---------------|-------|
| Successful    | 950   |
| Rate Limited  | 30    |
| Error (500s)  | 20    |
| Timeout      | 0     |

Total Requests = 1000

This table shows how many requests succeeded, were blocked by rate limits, or failed due to errors.
    

Precision vs Recall tradeoff with concrete examples

In error handling and rate limits, the tradeoff is between strict limits and user experience. Setting very low rate limits reduces errors but may block legitimate users (false positives). Setting high limits improves access but risks system overload (false negatives).

Example: A chat app with strict rate limits may block users sending many messages quickly (high precision in blocking bad requests) but annoy fast users (low recall of good requests). A looser limit improves recall but risks slowdowns.

What "good" vs "bad" metric values look like for this use case

• Good: Error rate under 1%, rate limit triggered only on abuse, latency under 200ms, throughput meets demand.
• Bad: Error rate above 5%, frequent rate limit blocks for normal users, latency spikes over 1 second, system crashes under load.

Metrics pitfalls

• Ignoring error types: Treating all errors equally hides critical failures.
• Data leakage: Not separating test and production logs can mislead error rates.
• Overfitting to metrics: Tuning only to reduce error rate may cause overly strict rate limits harming users.
• Accuracy paradox: High success rate may hide many blocked users if rate limits are too strict.

Self-check question

Your system shows 98% success rate but 12% of legitimate users get blocked by rate limits. Is it good for production? Why or why not?

Answer: No, because even though most requests succeed, blocking 12% of good users harms user experience and may reduce trust. Rate limits need adjustment to balance protection and access.

Key Result

Error rate, latency, and rate limit triggers are key metrics to balance system reliability and user experience.

Practice

1. What is the main purpose of using error handling in AI applications?

easy

A. To keep the app running smoothly even when problems happen

B. To speed up the AI model training process

C. To increase the number of requests sent to the server

D. To reduce the size of the AI model

Solution

1. Step 1: Understand error handling purpose

   Error handling is used to manage unexpected problems during app execution.

2. Step 2: Connect to AI app context

   In AI apps, error handling helps keep the app running smoothly despite issues.

3. Final Answer:

   To keep the app running smoothly even when problems happen -> Option A

4. Quick Check:

   Error handling = keep app running smoothly [OK]

Hint: Error handling means catching problems to avoid crashes [OK]

Common Mistakes:

• Thinking error handling speeds up training
• Confusing error handling with increasing requests
• Believing error handling reduces model size

2. Which Python syntax correctly catches an error when calling an AI API?

easy

A. try:
response = call_api()
except:
print('Error occurred')

B. catch:
response = call_api()
try:
print('Error occurred')

C. if error:
response = call_api()
else:
print('Error occurred')

D. error handling:
response = call_api()
except:
print('Error occurred')

Solution

1. Step 1: Identify correct try-except syntax

   Python uses try: block followed by except: to catch errors.

2. Step 2: Match syntax with options

   try: response = call_api() except: print('Error occurred') uses correct try-except structure; others use invalid keywords.

3. Final Answer:

   try:\n response = call_api()\nexcept:\n print('Error occurred') -> Option A

4. Quick Check:

   try-except syntax = try: response = call_api() except: print('Error occurred') [OK]

Hint: Remember Python uses try: and except: blocks [OK]

Common Mistakes:

• Using catch instead of except
• Using if error instead of try-except
• Writing invalid keywords like error handling:

3. What will the following Python code print if the API returns a rate limit error?

import time

try:
    response = call_api()
except RateLimitError:
    print('Rate limit hit, waiting...')
    time.sleep(10)
    response = call_api()
print('Done')

medium

A. Error: RateLimitError not caught

B. Done

C. Rate limit hit, waiting...

D. Rate limit hit, waiting...\nDone

Solution

1. Step 1: Understand try-except with RateLimitError

   If call_api() raises RateLimitError, except block runs printing message and waits 10 seconds.

2. Step 2: After waiting, call_api() runs again and then prints 'Done'

   So output includes the message and 'Done' on separate lines.

3. Final Answer:

   Rate limit hit, waiting...\nDone -> Option D

4. Quick Check:

   RateLimitError caught, message + Done printed [OK]

Hint: Exception caught prints message then continues [OK]

Common Mistakes:

• Assuming no message prints
• Thinking program crashes on rate limit
• Ignoring the second call_api() after sleep

4. Identify the error in this code snippet handling rate limits:

try:
    response = call_api()
except RateLimitError
    print('Too many requests')
    time.sleep(5)
    response = call_api()

medium

A. call_api() should not be retried

B. time.sleep() cannot be used in except block

C. Missing colon after except RateLimitError

D. print statement syntax is incorrect

Solution

1. Step 1: Check except syntax

   Python requires a colon ':' after except RateLimitError to start the block.

2. Step 2: Verify other parts

   time.sleep() is valid, retrying call_api() is allowed, print syntax is correct.

3. Final Answer:

   Missing colon after except RateLimitError -> Option C

4. Quick Check:

   except needs colon ':' [OK]

Hint: except lines always end with a colon ':' [OK]

Common Mistakes:

• Forgetting colon after except
• Thinking sleep() is invalid in except
• Believing retry is not allowed

5. You want to build an AI app that calls an API but respects rate limits by retrying after waiting. Which code snippet correctly implements this with error handling and exponential backoff?

hard

A. import time
wait = 1
for _ in range(3):
try:
response = call_api()
break
except RateLimitError:
time.sleep(wait)
wait *= 2

B. import time
wait = 1
while True:
try:
response = call_api()
break
except RateLimitError:
time.sleep(wait)
wait *= 2

C. import time
wait = 1
for _ in range(3):
try:
response = call_api()
except RateLimitError:
wait *= 2
time.sleep(wait)
else:
break

D. import time
wait = 1
while True:
try:
response = call_api()
except RateLimitError:
time.sleep(wait)
wait += 1
else:
break

Solution

1. Step 1: Understand exponential backoff with retries

   We want to retry after waiting, doubling wait time each failure, and stop on success.

2. Step 2: Analyze options for correct loop and break

   import time wait = 1 while True: try: response = call_api() break except RateLimitError: time.sleep(wait) wait *= 2 uses while True loop, tries call_api(), breaks on success, and doubles wait after RateLimitError.

3. Step 3: Check other options

   import time wait = 1 for _ in range(3): try: response = call_api() break except RateLimitError: time.sleep(wait) wait *= 2 breaks on success but uses for loop with fixed tries (less flexible). import time wait = 1 while True: try: response = call_api() except RateLimitError: time.sleep(wait) wait += 1 else: break increments wait linearly, not exponential. import time wait = 1 for _ in range(3): try: response = call_api() except RateLimitError: wait *= 2 time.sleep(wait) else: break doubles wait before sleep, but order is less clear.

4. Final Answer:

   import time wait = 1 while True: try: response = call_api() break except RateLimitError: time.sleep(wait) wait *= 2 -> Option B

5. Quick Check:

   Retry loop with exponential backoff = import time wait = 1 while True: try: response = call_api() break except RateLimitError: time.sleep(wait) wait *= 2 [OK]

Hint: Use while True with break and double wait after error [OK]

Common Mistakes:

• Using for loop limits retries too strictly
• Incrementing wait linearly instead of doubling
• Not breaking loop on success