Mock Exam

(Last updated: Mar 15, 2024)

Mock Final Exam

Below we provide a mock final exam and the answers.

Mock Mid-term Exam

Notice that this mock exam only contains a part of the real exam. In the real exam, there will be more questions.

We have defined several terms in the table below. You can refer back to them when the questions mention these terms. The feature (i.e., the input) is $x$. The ground truth (i.e., the true output) is $y$. The predicted output is $z$.

Term	Equation
Identity activation function $f(x)$	$f(x)=x$
Sigmoid activation function $f(x)$	$f(x)=1/(1+e^{-x})$
Tanh activation function $f(x)$	$f(x)=(e^{x}-e^{-x})/(e^{x}+e^{-x})$
Hinge loss function $f(y,z)$	$f(y,z)=\max(1-yz,0)$
Squared error loss function $f(y,z)$	$f(y,z)=(y-z)^2$
Perceptron loss function $f(y,z)$	$f(y,z)=\max(-yz,0)$
Logistic loss function $f(y,z)$	$f(y,z)=\log(1+e^{-yz})$
Binary cross-entropy loss function $f(y,z)$	$f(y,z)=(1-y)\log_{2}(1-z)-y\log_{2}(z)$
Softmax activation function $f(x)$, where $x_i$ means the $i^{th}$ element in array $x$, and $n$ means the total number of elements in array $x$	$f(x_i) = e^{x_i}/\sum_{j=1}^{n}e^{x_j}$
Entropy H for a coin with two sides (one side has probability $p_{1}$, and another side has probability $p_{2}$)	$H = p_{1}\log_{2}(1/p_{1}) + p_{2}\log_{2}(1/p_{2})$

Q1: Which of the following statements about missing data is FALSE?

Missing Completely At Random (MCAR) means that the missing data is a completely random subset of the entire dataset.

Missing data can be treated by replacing them with a constant value.

Missing Not At Random (MNAR) cannot be solved simply with imputation.

Missing at Random (MAR) means the missing data is related to the variable that has the missing data.

Q2: Which of the following statements about overfitting is FALSE?

Overfitting means fitting the test data extremely well.

Overfitting usually happens when using a very complex model.

We cannot deal with overfitting by simply removing outliers.

Errors of the model that we trained can be decomposed into bias, variance, and noise. Overfitting usually happens when we train a model with high variance.

Q3: Which of the following statements about the Decision Tree model is TRUE? Notice that we refer to the general Decision Tree, not a specific implementation.

Decision Tree can be trained well using entropy as the node-splitting strategy.

When splitting nodes, Decision Tree cannot use the same feature multiple times.

Decision Tree cannot be used for the regression task.

Decision Tree can only handle categorical features.

Q4: The research team in a municipality wants to use data science to explain how air pollution affects citizens’ living quality in the city. The municipality requires that the model’s decision-making process is as transparent as possible. This means that a researcher can check the model and explain how the model converts an input to an output intuitively. Which of the following algorithms does NOT suit the purpose?

Decision Tree

Linear Regression

Logistic regression

Deep Neural Network

Q5: Suppose we flipped a coin 12 times, and we got 10 heads and 2 tails. What is the entropy of the result in this coin-flipping experiment?

0.17

0.20

0.65

0.83

Q6: Suppose we have an array A of data with 3 columns, and we want to put the data in a pandas data frame. We want to give the first column with name “C1”, the second column with name “C2”, and the third column with name “C3”. Which of the following code produces the desired output for array A (considering that we already imported the pandas package)? For example, the output should look like the following data frame:

	C1	C2	C3
0	5	21.6	100
1	11	48.3	213
2	1	44.2	433

pandas.DataFrame(A, rows=["C1", "C2", "C3"])

pandas.DataFrame(A, columns=["C1", "C2", "C3"])

pandas.DataFrame(A).set_index(["C1", "C2", "C3"])

pandas.DataFrame(A, columns=["C1", "C2", "C3"]).set_index(["C1"])

Q7: Suppose we have a pandas data frame D with 100 rows and one column C1. Column C1 has 25% missing data. We want to sum up all valid items in column C1. Which of the following code produces the desired output? For example, if D looks like the data frame below, the code should output 27, which is a sum of 3, 4, and 20.

	C1
0	NaN
1	3
2	4
3	20

D.drop("C1", axis=1).sum()

D.dropna().sum()["C1"]

D.sum("C1")

D.groupby("C1").sum()

Q8: Which of the following about data modeling is TRUE?

R-squared is a common evaluation metric for classification models.

We can just use the training data to evaluate if the model will work well.

Classification is used to predict categorical labels, and regression is used to predict continuous values.

When using Random Forest, we can usually put raw data into the model without feature engineering.

Q9: Which of the following is the order that best describes the process of training a Decision Tree classifier?

A: Split nodes iteratively using an optimization algorithm (e.g., gradient descent) with the information gain criterion
B: Split nodes iteratively based on the information gain criterion until all leaf nodes are pure (i.e., all data has only one class) or a stopping condition is met
C: Choose the best feature using the information gain criterion for the root node
D: Evaluate the classifier’s performance using precision, recall, and F1 score

CAD

CBD

ACBD

CBAD

Q10: Which of the following practices is recommended in the data science pipeline?

Visualizing and exploring data in addition to using descriptive statistics

Sticking to a linear data science pipeline that starts from problem framing and data preparation to model building

Assuming that someone else has already framed the data science problem

Using the same modeling technique to deal with all different types of data

Q11: Given a pandas.DataFrame D, which of the following best explains the purpose of the following Python Pandas code?

D[(D["c1"]>=3)&(D["c1"]<=5)]

The code calculates the sum of the values in column “c1” that are greater than or equal to 3 and less than or equal to 5.

The code modifies the values in column “c1” to be between 3 and 5 if they have other values.

The code creates a new column in DataFrame D that flags rows with “c1” values between 3 and 5 with True, and False otherwise.

The code filters the DataFrame D to include only the rows where the values in column “c1” are between 3 and 5 (including 3 and 5).

Answers

Click to show the answers
- Q1: 4
- Q2: 1
- Q3: 1
- Q4: 4
- Q5: 3
- Q6: 2
- Q7: 2
- Q8: 3
- Q9: 2
- Q10: 1
- Q11: 4

Term	Equation
Identity activation function \(f(x)\)	\(f(x)=x\)
Sigmoid activation function \(f(x)\)	\(f(x)=1/(1+e^{-x})\)
Tanh activation function \(f(x)\)	\(f(x)=(e^{x}-e^{-x})/(e^{x}+e^{-x})\)
Hinge loss function \(f(y,z)\)	\(f(y,z)=\max(1-yz,0)\)
Squared error loss function \(f(y,z)\)	\(f(y,z)=(y-z)^2\)
Perceptron loss function \(f(y,z)\)	\(f(y,z)=\max(-yz,0)\)
Logistic loss function \(f(y,z)\)	\(f(y,z)=\log(1+e^{-yz})\)
Binary cross-entropy loss function \(f(y,z)\)	\(f(y,z)=(1-y)\log_{2}(1-z)-y\log_{2}(z)\)
Softmax activation function \(f(x)\), where \(x_i\) means the \(i^{th}\) element in array \(x\), and $n$ means the total number of elements in array \(x\)	\(f(x_i) = e^{x_i}/\sum_{j=1}^{n}e^{x_j}\)
Entropy H for a coin with two sides (one side has probability \(p_{1}\), and another side has probability \(p_{2}\))	\(H = p_{1}\log_{2}(1/p_{1}) + p_{2}\log_{2}(1/p_{2})\)