Training vs testing data quiz

Question 1

What is training data in machine learning?

Accepted Answer

The examples a model learns patterns from

Answer

The final score of a model

Answer

A type of computer chip

Answer

The internet connection used by the model

Question 2

What is testing data used for?

Accepted Answer

To check how well the model learned on unseen examples

Answer

To teach the model new patterns

Answer

To delete the model

Answer

To make the model run faster

Question 3

Why should training data and testing data be kept separate?

Accepted Answer

To get an honest measure of performance on new data

Answer

To save computer memory

Answer

To make training faster

Answer

Because the law requires it

Question 4

A common way to split a dataset is 80% for training and 20% for testing. What does this mean?

Accepted Answer

80% for training, 20% for testing

Answer

80% for testing, 20% for training

Answer

All data for training only

Answer

All data for testing only

Question 5

What does it mean when a model 'memorizes' the training data instead of learning?

Accepted Answer

Overfitting

Answer

Underfitting

Answer

Data cleaning

Answer

Cross-checking

Question 6

A model scores 99% on training data but only 60% on testing data. What is most likely happening?

Accepted Answer

Overfitting

Answer

Underfitting

Answer

Perfect learning

Answer

No training happened

Question 7

What is overfitting in simple terms?

Accepted Answer

Learning training data too well including noise

Answer

Learning too little from data

Answer

Using too much testing data

Answer

Running out of memory

Question 8

What is underfitting?

Accepted Answer

Model too simple to learn patterns

Answer

Model too complex for the data

Answer

Model with perfect accuracy

Answer

Model with too much data

Question 9

Why is testing on unseen data compared to a final exam?

Accepted Answer

Because it checks understanding on unseen questions

Answer

Because it is graded by a teacher

Answer

Because it happens in a classroom

Answer

Because it is always multiple choice

Question 10

What is 'data leakage' between training and testing sets?

Accepted Answer

When test information sneaks into training

Answer

When data is lost from the disk

Answer

When the model runs too slowly

Answer

When data is encrypted

Question 11

Why is it useful to shuffle data before splitting it into training and testing sets?

Accepted Answer

To ensure both sets get a fair mix of examples

Answer

To make the file smaller

Answer

To delete duplicate data

Answer

To speed up the computer

Question 12

What is a validation set, and how is it different from the testing set?

Accepted Answer

It is used to tune the model before the final test

Answer

It is the same as the testing set

Answer

It is only used after deployment

Answer

It is the training data renamed

Question 13

If you train a model to recognize handwritten digits, what would good training data include?

Accepted Answer

Many varied labeled images of digits

Answer

Only one perfect image of each digit

Answer

Pictures of animals

Answer

Random text without labels

Question 14

Why does using more high-quality training data usually help a model?

Accepted Answer

It exposes the model to more situations to learn from

Answer

It always slows learning down

Answer

It makes the model memorize faster

Answer

It removes the need for testing

Question 15

What does 'generalization' mean for a machine learning model?

Accepted Answer

Performing well on new unseen data

Answer

Memorizing the training data exactly

Answer

Running on many computers

Answer

Using a large file size

Question 16

In Python with scikit-learn, which function is commonly used to split data into training and testing sets?

Accepted Answer

train_test_split()

Answer

split_data()

Answer

make_model()

Answer

read_csv()

Question 17

In train_test_split, what does the test_size=0.2 setting mean?

Accepted Answer

Use 20% of data for testing

Answer

Use 2 examples for testing

Answer

Train for 0.2 seconds

Answer

Test 20 times

Question 18

Why might setting random_state to a fixed number in train_test_split be helpful?

Accepted Answer

It makes the data split repeatable each run

Answer

It makes the model more accurate

Answer

It deletes random data

Answer

It encrypts the dataset

Question 19

What is cross-validation in simple terms?

Accepted Answer

Repeatedly testing using different parts of the data

Answer

Training on test data only

Answer

Mixing two different models

Answer

Deleting half the data

Question 20

Why should the testing data come from the same kind of source as real-world use?

Accepted Answer

So it matches real situations the model will face

Answer

To make testing faster

Answer

Because it is cheaper

Answer

So the model can memorize it

Question 21

A student trains a spam filter and tests it on the exact same emails used for training. Why is the high accuracy misleading?

Accepted Answer

The model already saw those emails, so it just recalls them

Answer

The model is genuinely perfect

Answer

The emails were too short

Answer

Accuracy is never reliable

Question 22

What is the role of labels in supervised training data?

Accepted Answer

They are the correct answers the model learns from

Answer

They make files larger

Answer

They speed up the computer

Answer

They are random noise

Question 23

How can biased training data lead to an unfair model?

Accepted Answer

The model can learn imbalances and treat groups unfairly

Answer

Biased data has no effect on fairness

Answer

It only affects testing speed

Answer

It makes the model more accurate for everyone

Question 24

Why is it important not to peek at the testing data while building a model?

Accepted Answer

It keeps the final evaluation honest and unbiased

Answer

Peeking makes the model train faster

Answer

It saves disk space

Answer

It is required by Python

Question 25

If a model performs poorly on both training and testing data, what is likely the problem?

Accepted Answer

Underfitting

Answer

Overfitting

Answer

Data leakage

Answer

Perfect generalization

Question 26

What does 'unseen data' mean in machine learning?

Accepted Answer

Data the model has never encountered before

Answer

Data the model trained on

Answer

Data with no labels

Answer

Data stored in the cloud

Question 27

Why might a tiny testing set give an unreliable performance score?

Accepted Answer

A few results can swing the score, making it unstable

Answer

Small test sets are always best

Answer

It makes training impossible

Answer

It guarantees overfitting

Question 28

What is the main goal of training a machine learning model?

Accepted Answer

To learn patterns that work on new data

Answer

To memorize every example exactly

Answer

To delete the testing data

Answer

To run as slowly as possible

Question 29

How is studying for a test similar to a model using training and testing data?

Accepted Answer

Practice problems are like training and the exam is like testing

Answer

Studying is unrelated to machine learning

Answer

Exams are like training data

Answer

There is no similarity

Question 30

Why can a model that scores well on testing data still fail in the real world?

Accepted Answer

Real conditions can differ from the test data over time

Answer

Testing scores guarantee real-world success

Answer

The model forgets its training instantly

Answer

Testing always uses fake data

Test yourself