wayground logo

Free Printable Worksheets

Font size

S
M
L
XL
Worksheets

Worksheet MCQs extracted from provided images

Total questions: 125

Worksheet time: 1hrs 3mins

Name
Class
Date
1.

Which function best describes the size of the instance space with k attributes, each with n

a)

k*n

b)

nkn^k

c)

knk^n

d)

log2(n)

2.

Which method helps merge with fuzzy logic or approximate match?

a)

merge_asof()

b)

merge_fuzzy()

c)

concat()

d)

align()

3.

If a DNN uses only linear activations in all layers, it is equivalent to:

a)

SVM

b)

One-layer network

c)

Linear mapping

d)

Nonlinear classifier

4.

Merging two DataFrames on a common column uses which function?

a)

merge

b)

concat

c)

append

d)

join

5.

Doubling the distance between input and center in a Gaussian RBF affects activation how?

a)

Doubles

b)

Unchanged

c)

Decreases exponentially

d)

Increases linearly

6.

Which method assigns more weight to closer neighbors?

a)

Decision Tree

b)

Naïve Bayes

c)

Locally Weighted

d)

Bagging

7.

Which method helps merge with fuzzy logic or approximate match?

a)

merge_asof()

b)

merge_fuzzy()

c)

concat()

d)

align()

8.

Which of these techniques directly incorporates inductive bias?

a)

Random initialization

b)

Hypothesis pruning

c)

Uniform sampling

d)

High variance tuning

9.

What happens if every new example contradicts existing S and G sets in Candidate Elimination?

a)

Version space enlarges

b)

Version space

c)

G generalizes

d)

S becomes G

10.

What determines smoothness of the RBF decision boundary?

a)

Distance function

b)

Kernel width (σ)

c)

Learning rate

d)

Number of classes

11.

In a dataset, a feature causes splits [10,10], [15,5]. What is the weighted Gini index?

a)

0.5

b)

0.48

c)

0.45

d)

0.52

12.

Which of the following is a probabilistic model?

a)

Naïve Bayes

b)

Decision Tree

c)

SVM

d)

KNN

13.

The decision boundary in KNN is:

a)

Always linear

b)

Always nonlinear

c)

Can be complex

d)

Always fixed

14.

Layer outputs: [1.5, -0.5, 0.0], ReLU output = ?

a)

[1.5, 0, 0]

b)

[1.5, -0.5, 0.0]

c)

[1.5, 0.5, 0]

d)

[0, 0, 0]

15.

The Bayes Optimal Classifier aims to:

a)

Minimize computation

b)

Maximize accuracy

c)

Use all hypotheses

d)

Reduce variance

16.

Heuristic search is used in decision trees to:

a)

Optimize depth

b)

Select best attribute

c)

Normalize output

d)

Encode inputs

17.

A ReLU activation function outputs what for an input of -2.5?

a)

-2.5

b)

0

c)

2.5

d)

1

18.

Which of the following best describes G boundary behavior upon receiving a consistent input?

a)

Becomes specific

b)

Gets eliminated

c)

Becomes more general

d)

Refines by specialization

19.

Radial Basis Function (RBF) is used in:

a)

Decision Trees

b)

SVM

c)

Neural Networks

d)

Case-Based Learning

20.

Which of the following is not a metric distance function?

a)

Manhattan

b)

Cosine similarity

c)

Euclidean

d)

Mahalanobis

21.

The learner’s task in a supervised learning setup is to:

a)

Find the input space

b)

Find the hypothesis

c)

Design the test set

d)

Create labels

22.

For Naïve Bayes: P(Sunny)=0.6, P(Play=Yes|Sunny)=0.8. Compute joint.

a)

0.48

b)

0.36

c)

0.72

d)

0.64

23.

Dataset = 1000 points, k=10, distance calculation is O(d). What is per-query complexity?

a)

O(50000)

b)

O(10000)

c)

O(500)

d)

O(10)

24.

A perceptron can solve which type of problems?

a)

Non-linear

b)

Linear separable

c)

All ML problems

d)

Polynomial only

25.

What is the primary goal of machine learning?

a)

Predicting data

b)

Extracting rules

c)

Learning from data

d)

Cleaning data

26.

Which method stacks columns into rows?

a)

melt()

b)

pivot()

c)

groupby()

d)

explode()

27.

If arr = np.array([[1,2],[3,4],[5,6]]), what is arr[:,0]?

a)

[1 3 5]

b)

[2 4 6]

c)

[1 2 3]

d)

Error

28.

Which has higher expected accuracy: BOC or Gibbs?

a)

Always Gibbs

b)

Always BOC

c)

Depends on prior

d)

Equal

29.

For 3 models: A(-120,4p), B(-100,6p), C(-110,5p), which has better AIC?

a)

A

b)

B

c)

C

d)

All same

30.

Given x = np.array([[1,2],[3,4]]), what is x[0][1]?

a)

2

b)

1

c)

3

d)

Error

31.

If mutation rate = 0.01, how many mutations occur in population of 100 genes?

a)

1

b)

10

c)

100

d)

0.1

32.

Version spaces are updated when:

a)

A new attribute is

b)

A new instance is

c)

Output is generated

d)

Error occurs

33.

An RBF network uses a Gaussian function with variance = 1. What is the output for input x = 1?

a)

eλ−0.5e^{\lambda} - 0.5

b)

e−1e^{-1}

c)

e−2e^{-2}

d)

e0e^0

34.

In machine learning, the target function refers to:

a)

Learned weights

b)

Mapping being

c)

Ground truth

d)

Loss function

35.

The output in KNN is predicted using:

a)

Training loss

b)

Aggregated neighbor

c)

Distance matrix

d)

Model weights

36.

You observe data x=[2,2,1], with latent z having P(z=1)=0.7, compute E[z] over dataset.

a)

2.1

b)

1.5

c)

2.8

d)

3

37.

A network with only linear activation functions is equivalent to:

a)

Deep network

b)

Single linear transformation

c)

PCA

d)

SVM

38.

What is inductive learning?

a)

Drawing specific

b)

Drawing general

c)

Drawing from noise

d)

Drawing images from text

39.

A test point’s 7 nearest neighbors have distances: [0.3, 0.5, 0.7, 1.0, 1.1, 1.3, 1.8] and labels A or B. Which class wins the vote?

a)

A

b)

B

c)

Tie

d)

Cannot be computed

40.

Given a dataset of 10 i.i.d. samples with likelihood L=0.6, what is the log-likelihood?

a)

-5.11

b)

-6.4

c)

-8.2

d)

-10

41.

A version space has 32 hypotheses. After seeing 2 consistent examples, only 8 remain. How many were eliminated?

a)

8

b)

16

c)

24

d)

28

42.

Which layer reduces input dimensionality in CNN-based deep nets?

a)

ReLU

b)

Fully connected

c)

Pooling

d)

Batch norm

43.

Backpropagation algorithm relies heavily on:

a)

Chain Rule

b)

Matrix inversion

c)

Dropout

d)

Normalization

44.

np.savez('data.npz', x=[1,2], y=[3,4]) creates a file that stores:

a)

Multiple arrays

b)

Single array

c)

Dictionary

d)

JSON structure

45.

In EM algorithm, if E-step results in expected value of latent variable Z = 0.7, and M-step is performed, what is updated?

a)

Model parameters

b)

Likelihood

c)

Gradient

d)

Prior

46.

Hypothesis space search in GA involves:

a)

Searching weights

b)

Random walk

c)

Chromosome evolution

d)

Gradient updates

47.

Forward pass through a network is used to:

a)

Compute accuracy

b)

Update weights

c)

Generate predictions

d)

Normalize data

48.

Which aggregation function gives cumulative sum of a NumPy array?

a)

np.cumsum()

b)

np.sum()

c)

np.accumulate()

d)

np.add()

49.

You run 10-fold cross-validation on a dataset of 900 points using KNN. How many samples are used for validation in each fold?

a)

90

b)

810

c)

100

d)

450

50.

In noisy domains, which inductive bias is likely to generalize better?

a)

Simplicity bias

b)

Robustness bias

c)

Overfitting bias

d)

Exhaustive bias

51.

Neural networks adjust weights based on:

a)

Learning rule

b)

Data similarity

c)

Fixed values

d)

Decision boundaries

52.

What is the major drawback of Euclidean distance in high-dimensional spaces?

a)

Too slow

b)

Loses discriminativity

c)

Works only with integers

d)

Requires normalization

53.

Which operation is NOT part of genetic algorithm?

a)

Mutation

b)

Crossover

c)

Selection

d)

Backpropagation

54.

A small value of K in KNN may lead to:

a)

Underfitting

b)

High bias

c)

Overfitting

d)

High accuracy

55.

In backpropagation, gradient is 0.3, learning rate is 0.2. Weight update = ?

a)

-0.06

b)

0.06

c)

0.1

d)

-0.1

56.

Which factor is most critical in distinguishing supervised from unsupervised learning?

a)

Number of features

b)

Labeled data

c)

Training speed

d)

Number of iterations

57.

Why is activation function necessary in multilayer perceptron?

a)

For non-linearity

b)

To speed training

c)

Reduce weights

d)

Make matrix invertible

58.

Which method merges on index instead of columns?

a)

join()

b)

merge()

c)

concat()

d)

groupby()

59.

Retention in CBR refers to:

a)

Forgetting bad cases

b)

Adding all failed cases

c)

Storing new experience

d)

Learning rate tuning

60.

LWR belongs to which type of machine learning algorithm family?

a)

Instance-based

b)

Reinforcement

c)

Bayesian

d)

Generative

61.

Given P(C)=0.5, P(x1|C)=0.8, P(x2|C)=0.7, compute P(x1,x2,C) for Naïve Bayes.

a)

0.28

b)

0.35

c)

0.2

d)

0.45

62.

Given P(A)=0.4, P(B)=0.5, and P(B|A)=0.7, compute P(A*B).

a)

0.35

b)

0.28

c)

0.2

d)

0.25

63.

In a dataset of 500 cases, 100 are invalid. How many valid cases are available for retrieval?

a)

300

b)

400

c)

450

d)

350

64.

If all features are independent given class, then Naïve Bayes performs like:

a)

Logistic Regression

b)

Decision Tree

c)

Gibbs

d)

Bayes Optimal

65.

If the output error δ=0.4 and input x=3, with η=0.05, calculate Δw.

a)

0.06

b)

0.03

c)

0.08

d)

0.05

66.

What is the Manhattan distance between (2, 4) and (5, 8)?

a)

5

b)

7

c)

6

d)

4

67.

If x = np.array([1,2,3]), what is x + x[::-1]?

a)

[4 4 4]

b)

[2 4 6]

c)

[3 4 4]

d)

Error

68.

Gibbs Algorithm selects hypotheses based on:

a)

Randomization

b)

Distance

c)

Posterior probabilities

d)

Time complexity

69.

Perceptron weights are [1, -1], input is [2, 2], threshold is 0. What is the output?

a)

0

b)

1

c)

2

d)

-2

70.

A test point lies at distances [1.2, 1.8, 2.0] from 3 neighbors (classes A, B, A). Using inverse

a)

A

b)

B

c)

C

d)

Cannot determine

71.

In which case would weight update be minimal in backpropagation?

a)

Large error

b)

Small gradient

c)

High input value

d)

High learning rate

72.

The main advantage of case-based reasoning is:

a)

Requires no training

b)

Fast prediction

c)

Generalizes poorly

d)

Learns without data

73.

If BOC combines 3 hypotheses with individual errors 0.2, 0.3, 0.4, what is the best-case BOC

a)

< 0.2

b)

0.3

c)

0.25

d)

0.4

74.

What does np.random.seed(42) ensure?

a)

Same output every time

b)

Faster random

c)

GPU support

d)

Parallel sampling

75.

Which condition leads to the collapse of version space?

a)

No positive example

b)

Redundant attributes

c)

Conflicting examples

d)

Too many generalizations

76.

The general-to-specific ordering of hypotheses is based on:

a)

Accuracy

b)

Coverage

c)

Specificity

d)

Generalization

77.

LWR on 8000 samples takes 0.8 seconds to predict one output. What’s the estimated time

a)

80 sec

b)

160 sec

c)

0.16 sec

d)

200 sec

78.

A system evaluates 60 cases and selects 3 top matches. How many were ignored in the final

a)

57

b)

3

c)

30

d)

33

79.

You have 3 features with information gains: A=0.2, B=0.3, C=0.1. Which one is selected by

a)

A

b)

B

c)

C

d)

Any

80.

Update rule for perceptron adjusts weights only when:

a)

Correctly classified

b)

Output = 1

c)

Misclassified

d)

Always

81.

Deep networks are better than shallow ones because:

a)

Easier to train

b)

Need less data

c)

Learn complex functions

d)

Use few neurons

82.

What happens if crossover probability is set to 0?

a)

No offspring

b)

Max mutation

c)

Constant parents

d)

All solutions same

83.

A 3-layer DNN with 100, 50, and 10 neurons respectively has how many parameters (weights

a)

3050

b)

3020

c)

2500

d)

4010

84.

Which of the following is NOT a genetic operator?

a)

Selection

b)

Evaluation

c)

Mutation

d)

Crossover

85.

An RBF network has three centers: c1 = [1,1], c2 = [4,4], c3 = [7,7]. Input x = [4,4]. Which center

a)

c1

b)

c2

c)

c3

d)

All Equal

86.

Which function in Pandas loads multiple sheets from Excel?

a)

pd.read_excel()

b)

pd.load_excel()

c)

pd.read_xls()

d)

pd.excel_load()

87.

Which Pandas method saves DataFrame in compressed .zip format?

a)

to_csv(compression=)

b)

to_zip()

c)

compress_csv()

d)

export_zip()

88.

What happens if you save a DataFrame using to_csv() without specifying index=False?

a)

Index is saved

b)

Only values are saved

c)

Header is missing

d)

File is corrupted

89.

What type of learning does not require labeled data?

a)

Reinforcement

b)

Supervised

c)

Semi-supervised

d)

Unsupervised

90.

Given an input [1, 0, 1] and weights [0.5, 0.2, 0.3], what is the neuron output before applying

a)

0.5

b)

0.8

c)

1

d)

0.3

91.

If an attribute has 4 possible values, how many minimal generalizations are there when

a)

1

b)

2

c)

3

d)

4

92.

In model selection, BIC prefers models with...

a)

High likelihood &

b)

Low likelihood

c)

Many params

d)

Small dataset

93.

In decision trees, which attribute is chosen at each node?

a)

One with smallest value

b)

Randomly selected

c)

One with highest information gain

d)

One with least support

94.

Given a = np.array([1,2,3,4,5]), what is a[a%2 == 0]?

a)

[2 4]

b)

[1 3 5]

c)

[0 2 4]

d)

[2 3 4]

95.

Decision tree pruning is used to:

a)

Add more branches

b)

Remove underfitting

c)

Reduce overfitting

d)

Increase tree height

96.

A learning algorithm that assumes feature independence follows what kind of bias?

a)

Interaction bias

b)

Independence bias

c)

Naïve bias

d)

Structural bias

97.

RBF networks are best suited for:

a)

Time series

b)

Classification with

c)

Image segmentation

d)

Reinforcement learning

98.

Which activation function is prone to vanishing gradients?

a)

ReLU

b)

Sigmoid

c)

Softmax

d)

Max Pool

99.

You roll a biased die: P(6)=0.3, P(1–5)=0.14 each. Given an even number, what's the probability?

a)

0.3

b)

0.35

c)

0.5

d)

0.21

100.

Why can MDL avoid overfitting?

a)

Adds noise

b)

Ignores likelihood

c)

Penalizes complexity

d)

Maximizes entropy

101.

What makes RBF networks different from standard MLPs?

a)

Linear activation

b)

Radial symmetry in

c)

Binary output

d)

Normalized input

102.

Dataset has 100 samples. Model A log-likelihood is -230, B is -200. Which has higher likelihood?

a)

Model A

b)

Model B

c)

Equal

d)

Cannot determine

103.

If P(h)=0.3, P(D|h)=0.5, compute P(h|D) assuming normalizer P(D)=0.5.

a)

0.3

b)

0.5

c)

0.15

d)

0.6

104.

Fitness function is used to:

a)

Train classifier

b)

Evaluate performance

c)

Combine layers

d)

Stop training

105.

What is the probability that Gibbs picks a correct hypothesis from H={h1,h2} with

a)

0.7

b)

0.5

c)

1

d)

0.3

106.

The depth of a decision tree indicates:

a)

Number of classes

b)

Size of dataset

c)

Number of attributes

d)

Complexity of decisions

107.

In a 1D Gaussian RBF ϕ(x)=exp⁡(−(x−c)2/2)\phi(x) = \exp(-(x - c)^2 / 2) , what is the output when x = 2 and c = 2?

a)

0

b)

0.5

c)

1

d)

e−1e^{-1}

108.

Given weights [0.2, 0.4] and inputs [1, -1], with bias 0.1, compute the perceptron output using

a)

0

b)

1

c)

-1

d)

0.5

109.

Chebyshev distance between (3,1) and (6,4) is:

a)

3

b)

2

c)

1

d)

5

110.

What file format is used by np.save()?

a)

.npy

b)

.npz

c)

.csv

d)

.txt

111.

What type of algorithm is backpropagation?

a)

Supervised

b)

Unsupervised

c)

Semi-supervised

d)

Reinforcement

112.

You use 1-NN on a noisy dataset and get unstable predictions. What’s a good fix?

a)

Use k=3

b)

Drop outliers

c)

Normalize again

d)

Add features

113.

Using sigmoid activation, what is the output if net input is 0.5? (approx.)

a)

0.73

b)

0.62

c)

0.5

d)

0.45

114.

When does increasing k in KNN result in higher misclassification on boundaries?

a)

When noisy

b)

Low-dimensional

c)

Minority near majority

d)

All of the above

115.

A gradient of 0 means:

a)

No weight change

b)

Full update

c)

Reset weights

d)

Faster convergence

116.

Which Pandas function is used to combine data along columns with alignment on indexes?

a)

join()

b)

concat()

c)

merge()

d)

groupby()

117.

P(A)=0.6, P(B|A)=0.5, P(B|¬A)=0.2. Find P(B).

a)

0.38

b)

0.42

c)

0.44

d)

0.36

118.

A real-time system must predict class within 5ms. KNN is used on a dataset with 10,000

a)

Accuracy low

b)

Not real-time

c)

High prediction latency

d)

Too simple

119.

Why does the Candidate Elimination algorithm fail with noisy labels?

a)

It has no bias

b)

It requires clustering

c)

It assumes consistency

d)

It lacks optimization

120.

If the hypothesis space has 2N2^N elements and one hypothesis is eliminated per example, how

a)

log⁡2N\log_2 N

b)

N

c)

2N2^N

d)

N/2

121.

A learning problem has 1000 examples and 10 candidate concepts. What is the probability of

a)

0.1

b)

1

c)

0.9

d)

0.01

122.

Which of these is valid for NumPy broadcasting?

a)

(3,1) + (1,4)

b)

(3,2) + (3,4)

c)

(5,5) + (5,)

d)

All except B

123.

Input x = [3,4], center c = [0,0]; compute RBF value with σ=5\sigma = 5 .

a)

e−1e^{-1}

b)

e−0.5e^{-0.5}

c)

e−0.32e^{-0.32}

d)

e−2e^{-2}

124.

How does inductive bias improve learning in high-dimensional spaces?

a)

Removes noise

b)

Reduces hypothesis

c)

Prevents generalization

d)

Avoids test error

125.

Which function is used to introduce non-linearity in neural networks?

a)

Weight function

b)

Bias function

c)

Activation function

d)

ReLU layer

Similar Resources on Wayground

EXAMEN DE FUNDAMENTOS DE PROGRAMACIÓN 1

121 questions

EXAMEN DE FUNDAMENTOS DE PROGRAMACIÓN 1

quiz

Worksheet

•

University

HÀNG HÓA CHƯƠNG 6

125 questions

HÀNG HÓA CHƯƠNG 6

quiz

Worksheet

•

University

Quiz_Review_Exam3

122 questions

Quiz_Review_Exam3

quiz

Worksheet

•

University

Compre

120 questions

Compre

quiz

Worksheet

•

University