Batch prediction vs real-time serving in MLOps - Practice Questions

Challenge - 5 Problems

🧠 Conceptual

intermediate

Key difference between batch prediction and real-time serving

Which statement best describes the main difference between batch prediction and real-time serving in machine learning?

A Batch prediction processes many data points at once periodically, while real-time serving predicts instantly for individual requests.

B Real-time serving only works with small datasets, batch prediction only works with large datasets.

C Batch prediction is always more accurate than real-time serving because it uses more data.

D Batch prediction requires manual input for each prediction, real-time serving automates all predictions.

💡 Hint

Think about how often predictions are made and how many data points are handled at once.

Batch prediction runs on large groups of data at scheduled times, while real-time serving responds immediately to single data inputs.

❓ Model Choice

intermediate

Choosing prediction method for fraud detection

You are building a fraud detection system that must flag suspicious transactions immediately as they happen. Which prediction method is best?

A Batch prediction, running once a day on all transactions.

B Real-time serving, predicting only after collecting 100 transactions.

C Batch prediction, running once a week on all transactions.

D Real-time serving, predicting on each transaction as it occurs.

💡 Hint

Consider how fast the system needs to respond to new data.

Real-time serving is needed to flag fraud instantly, batch prediction delays detection.

❓ Metrics

advanced

Evaluating latency in batch vs real-time prediction

You measure the average latency (time to get prediction) for batch prediction and real-time serving. Which is true?

A Batch prediction has higher latency per data point than real-time serving.

B Both have the same latency per data point.

C Real-time serving has higher latency per data point than batch prediction.

D Latency cannot be measured for batch prediction.

💡 Hint

Think about how predictions are processed individually or in groups.

Real-time serving predicts instantly but each prediction takes some time; batch prediction processes many points together, lowering average latency per point.

🔧 Debug

advanced

Identifying issue in real-time serving system

A real-time serving system is suddenly slow and sometimes fails to respond. Which is the most likely cause?

A The model used is too simple and cannot handle the data volume.

B The real-time serving system is overloaded with too many simultaneous requests.

C The batch prediction job is running too frequently and blocking resources.

D The data used for batch prediction is outdated.

💡 Hint

Consider what happens when many users request predictions at once.

Real-time serving can slow or fail if too many requests overload the system, causing delays or errors.

❓ Predict Output

expert

Output of batch vs real-time prediction code snippet

Given the code below, what is the output?

MLOps

def batch_predict(model, data):
    return [model(x) for x in data]

def real_time_predict(model, x):
    return model(x)

model = lambda x: x * 2
batch_data = [1, 2, 3]

batch_result = batch_predict(model, batch_data)
real_time_result = real_time_predict(model, 4)

print(batch_result, real_time_result)

A [2, 4, 6] 8

B [1, 2, 3] 4

C [2, 4, 6] [8]

D Error: model is not callable

💡 Hint

Look at how the model function is applied to data in batch and real-time.

The model doubles each input. Batch prediction applies it to a list, real-time applies it to one value.