MLOps & Model Deployment — Quiz | Data Science using Python

Question 1

Roughly how much of a real ML project is model training?

A About 90%

B About 10% — the rest is data, serving, monitoring, retraining

C 100%

D 50%

Question 2

Why do deployed models decay even if the code is unchanged?

A Bugs appear randomly

B The world drifts away from the training data

C Python updates

D Disks fill up

Question 3

Why save the whole Pipeline rather than just the estimator?

A It is smaller

B So serving uses the exact same preprocessing as training

C It trains faster

D It avoids Git

Question 4

Reproducing a model exactly requires versioning which four things?

A Only the code

B Code, data, model artefact, and environment

C Just the data and code

D Code and accuracy

Question 5

What does MLflow primarily do?

A Serve web pages

B Track experiments' parameters, metrics and artefacts (and register models)

C Clean data

D Replace Docker

Question 6

What is FastAPI used for here?

A Training models

B Wrapping a model in a validated web API endpoint

C Versioning data

D Plotting charts

Question 7

Where should you load the model in a serving app?

A Inside every request handler

B Once at startup

C Never — retrain each call

D In the Dockerfile

Question 8

What problem does Docker solve?

A Slow training

B 'Works on my machine' — it packages code+deps+runtime to run identically anywhere

C Missing data

D Overfitting

Question 9

What is data drift?

A A network outage

B The input distribution shifts away from training data

C The model file corrupting

D A coding bug

Question 10

What is concept drift?

A Inputs stay the same

B The relationship between inputs and the target changes

C The server restarts

D The dataset shrinks

Question 11

Which test can flag that a feature's distribution has shifted?

A A correlation

B A Kolmogorov-Smirnov (KS) two-sample test

C A confusion matrix

D A pivot table

Question 12

What does a mature, closed-loop MLOps system do?

A Trains once and stops

B Logs predictions, monitors metrics, alerts on drift, and retrains on fresh data

C Avoids monitoring

D Deletes old models immediately