Online self assessment for applicants of the Master programme Data Science TU Dortmund University Departments of Mathematics, Computer Science and Statistics Dear interested Data Seience students, ‘many thanks for your interest in our Master program. ‘* We would like to bring to your notice that our Master in Data Science is a consecutive Master program. ‘This means that our department offers both a Bachelor and a Master in Data Science, where the Master builds upon all the knowledge of our Bachelor. Please note that the Bachelor is taught in Germnan, while the Master js taught in English, ‘* Our Master program is open for carver changers with different backgrounds than our own Bachelor. However, wwe expect from you that you aze familiar with most of the topics that are taught in our Bachelor prograun. ‘This test is meant give you an overview of the most important contents of our Bachelor: Hence, a student who successfully finished our own Bachelor in Data Seicuce should be able to answer most of the question with case, ‘© You aro not expected to be able to answer all the questions. You should assess whether you have a sufficient mathematical background that you believe that with the corresponding effort you will be able to answer such questions, whether you are motivated to learn how to answer such questions and now concopts building upon the concepts addressed in these questions. ‘© While we require you to do the selfasessinent, we do not consider the answens yon actually gave when assessing your application. ‘This is not a knowledge test. ‘There will be no score, grade or points at the end. And your answers are not uscd to evaluate your application (you don't even have to send us your answers). ‘# This test is meant for you, for your own protection: If you join our Master courses, we expect you to know everything from our Bachelor. if you don't, you will not be able to follow our Master's courses appropriately, and it is rather likely the you will fail our program after investing a lot of time and money. ‘* Hence, we recommend that you to take this test seriously. It shows you what we expect from you, However, \we are not interested in how good you perform in this test. There will be no score, and instead we'll tell vou the correct solutions in the end. ‘© Tho test is designed in a multiple-choice way. For all 4 areas (Mathematics, Computer Science, Statisties and Data Seience), there will be several questions, a total of 50. For each question, there are multiple answors (mostly, 4 different: answers). You havo to decid for each answer whother it is right or wrong, '* After you finished the test, please fill ont our self-disclosure document. In this document: you have to sign ‘that you did the test. You have to upload this document on the uni-assist platform, Please double-check to sign it twice, once for the self-test and once for the report We wish yon good Inck!1 Mathematics ‘# Tho symbol In denotes the natural logarithm, that is, the logarithm with base e. * The symbols Z, Q, B, C denote the sets of integers, rational numbers, real numbers, and complex numbers, respectively: 1.1 Calenlus «) ey Question 1. For a €R, which statements do hold for the one-sided limit lim, a) The limit exists, ) Its value is 0. ©) Its value is 1 «) Its value is Question 2, Which statements do hold for the definite integral ["o™**sin-rde ? 1) The antiderivative is not explicitly caleulable. bb) The defini ntegral has a finite value, ©) Its value is ¢ ) Its value is 0. Question 3. ‘The following figure shows the graph of the derivative J" of a fauection f, where f is contimuons on the interval (0,4] and differentiable on the interval (0,4). Which of the following statements give the correct ordering of the function values (0), f(2), and f(4y? a) £0) < F() ») FA) < FQ) ©) F(2) 5 F(0) 4) FQ) = F(2) Question 4. Let f be the function defined by the series f(x) Le for all x such that 1 <2 <1. Which statements do hold?1a) The series converges absolutely. b) The derivative is f(r) = 7a ©) The derivative is f"(z) = 02". 4) Phe dexivative equals f(x) = 1.2. Linear Algebra Question 5. Consider the following system of lincar equations ar + y + 2 = 0 t+ yee ao : 2-0 ‘with solutions of the form (2, y,2) where 2,y, are real numbers. Which of the following statements are correct? 1a) The system is consistent. b) The sum of any two solutions is 1 solution. ©) The system has a unique solution, ) ‘Phe system has infinitely many solutions. Question 6. Which are eigenvalues of the matrix ( S ony a0 1.3. Analytic Geometry Question 7. Consider the solid in ryz-space, which contains all points (x, y,2) whose z-coordinate satislies Which statements do hold? 1) Tho solid is a sphere. ) ‘Phe solid is a pyramid, 6) Tes volmme is 87, lor 4) Hs sotmine is Question 8. Consider the fimetion g defined by 9(c, y) = e¥(y — 22) for all real x,y. Which of the following terms are needed to represent the length of the gradient Vg(1,—1) ? a) V10b) vB oe dr Question 9. A circular helix in xry2-space has the following parametric equations, where @ € R. 2() = Aeos0 y(@) = sind 20) = 30 Let (0) be the arclength of the helix from the point P(0) = (x(0),4(0), (0) to the point P(Q) = (4,0,0), and let D{G) be the distance between P(#) and the origin (0,0,0). Let £(8) = 10. Which statements do hold? a) O=4 b) 0=2 ©) To calculate the value of D for a given @, 2(9) and y(0) have to be evaluated explicitely: d) DO) = VB 1.4 Differential Equations Question 10. Let y: B+ be the real-valued function defined on the real line, which is the solution of the initial ‘value problem wa-ryts, — ¥(0)=2 Which statements are correct? 8) The problem is not uniquely solvable. b) The solu n y(2) contains an exponential function. ©) Him wz) =1 @) tin le) =0 2 Computer Science 2.1 Data Structures Question 11. The number of steps taken for searching the value x in a binary tree with m nodes .. 2) depends not on 2) depends on 1. ©) is Ollogy ). d) is O(log, n). Question 12. ‘The average-cise performance when looking up a single search ky 1) s better with a Linked List than with a Hash Table b) is better with a Hash Table than with an Array. ¢) is better with a Binary Search Tree than with a Hash Table 4) fs the same with a Linked List, an Array, and a Hash Table Question 13. Given 100000 numbers, the minimum height of a binary soarch troe that ean store all these numbers1) depends on the mmbers. ) is larger than 20 levels 6) is smaller than 19 levels. 4) ean be calculated as lo (100 000). Question 14. Wi ih of the following statements are correct for a max? 8) The toot always contains the lingest key b) All keys in the left subtree are always stualler than any key in the corresponding right subtree. ©) All leaves are located on the same level A) Bach subtree is also a mas-heap. Question 15. Which of the following statements nrc correct for a binary search tee? 18) The root always contains the largest key. ) Alleys in the lft subtree are always smaller than any key in the corresponding right subtree ©) All leaves are located on the same level dB 41 subtroe is also a binary search treo. Question 16. The following operations are applied to an empty stack «: push(1) push(2) push(3) pop) push (4) pop) ‘The result of a further s-popQ is a) amber bs) undefined o4 a2 2.2. Algorithms and Programming Question 17. Sorting a data set is an important sub-problem in data science. Given the size n of a data set, which statements are correct? 18) Bubble Sort has worst-case rim-time complexity O(n). }b) Bubble Sort has worst-case runtime complexity O(rlog(n)) ©) Bubble Sort has worst-case run-time complexity O(n?). d) Merge Sort has worst-case run-time complexity O(n). ©) Merge Sort has worst-case run-time complexity O(n log(n)) 4) Merge Sort has worst-case runctime complexity O(n?) 42) Quick Sort has worst-case run-time complexity O(n). 1h) Quiek Sort hnns worst-case run-time comple: ty Olwlog(n))-4) Quick Sort has worst-case run-time complexity O(n?) Question 18. CI and C2 are classes written in an object-oriented programming language (such as Java, Cf, or C++). Which of the following statements are correct if C1 is a superelass of C2? a) C1 is always an abstract elas 1) C2 contains all publie features defined by C1 «) Fach €2 object may be replaced by C1 object 4) C2is a subelass of C1 Question 19. The following function f uses recursion: def f(a): iene return n else return f(n-1) + £(n-2) Lot n be a valid inpnt, i.c., a natural mmber. Which of the following fmctions returns the same result but without a) def £(n) aco bed ifn=0 elsif n= 1 return b else for iin ion ccate ace bec revurn b b) dof £(m): aco while 1 > 0 acatit (it) return a ©) det £0) arr{0] <0 arr(i] <1 ifned return arr [a] else for iin 2. arr[i] < arr(i-t] + arr[i-2] return arr [a] () def £) arr{0..n] < [0, ifned return arr{n) else acofor 4 im O..n acat arf return a 2.3 Logic and Databases Question 20. If A, B, and C ure Boolean wisiables, which of the following statements are correct? a) AA(BVC) =(AAB)V (AAC) b) AV(BAC) =(AVB)A(AVC) ©) (AAB)VC=Cv(BAA) Question 21. A large retail company keeps sales data local to the individual brane ‘were performed. ‘To compute overall sales statisties, the company wants to avoid sending the full sales data set to 1, variance, mec, main) is sent from cach branch to the central site. Which of the following statements are correct? where sales transactions a central server. Instead, only aggregated sales information (sum, average, minin 1) The overall sm cam be derived from the sums per branch, 1) "The overall average ean be derived from the averages per branch ¢) The overall minimum can be derived from the minimums per branch. 4) ‘The overall variance can be derived from the varianoss per branch. 6) ‘The overall median ean be derived from the medians per branch £) ‘Tho overall imimm can be derived from the maximams per branch, Question 22. Consider the following table in a relational database. Last Name Ranke Shift Smith Manager 234 Morning Jones Custodian 33 Afternoon Smith Custodian 33 Evening Doe Clerical = 222 Morning According to the data shown in the table, which of the following comldl be candidate keys of the table? a) {Last Name} ) (Room) ©) {Shitty ) (Rank, Room} Shiit} ©) (Room, Question 23, ‘The database interface of a library allows searching only for a single attribute (such as “Tithe. oF -Author.) in each query. Your friend decided to extend it's functionality aud wrote an algorithm that allows searching for books that satisfy multiple predicates over single attributes in conjunction. He tells you the algorithm reuses the already implemented query functionality and works by intersecting the results (.book id's.) of queries ‘ower single attsibutes Which of the following assumptions ou your friend's algorithm are plausible? a) Its worst-case runctime necessarily increases exponentially with respect to the mumber of attributes in the query. ) ts worst-case run-time depends on the length of the longest result of the single-attibute queries. ©) Mt might be implemented using an join ) It might be implemented using sorting,2.4 Fundamentals of theoretical computer science Question 24. Given an implementation of an algorithm, you want to check formally its run-time performance before you apply the algorithm to big data sets, in order to prevent endless runs of algorithms on your computer. ‘Tho check if your algorithin runs endlessly on this data is depending on. 1) the length of the source code, itis a coding problem. ») function calls in the algorithan, it is a eall-graph problem, ©) recursion in the algorithm, i is software design problem. 4) the size of your data, i is a big data problem, Question 25. Which of the fllosng languages ane rogue? 4) Words that consist of only vowels (a, ‘e' ‘?, “ow. b) Words where the 6th-last character is a vowel. ©) Words that contain as many vowels as consonants (non-vorscls) 4d) Palindroms (reading the word backwords yields the same word). 2.5 Computer Architecture Question 26. In computer architecture, SIMD may refer to the situation whet. 1) multiple CPU cores can access the same memory concurrently. }) the same operation can be applied to multiple operands with only a single instruction ©) inultiple independent instruetions ean be exceted at the same time in the ssune CPU ore. d) multiple independent memory banks show up as a single address space. 3 Statistics 3.1 Descriptive Statistics Question 27. Which of the following sets have an arithmetic mean of 100, but a median smaller than 100? a) {80, 100, 120} b) (80, 80, 40} ©) {0, 50, 150} a) (60, 120, 120} Question 28, Can there be a set of data fitting to both the following histograms!” Which of these answers are correct?Histogram 1 Histogram 2 a) No, because the right one is calculated from positive data only. ») Yes, the right one includes all possible data from which the left one may be calenlated ©) No, the right one must be caleulated with at least one value greater than 8. «d) No, the left one ean not have been calculated with a valwe of 10 or more. Question 29. Caleulate esti ‘as well as the Pearson coeffit ates of the standard deviations s2, sy of the samples x = (5,9,7) and y= (—1,2,5) nt of correlation ray of 2 and y. Which of the following answers are correct? a) ey b) my =0 ©) te 25 4) roy = ©) try = 4 Question 30. Consider the following seatter plot.‘The coofficiont of correlation of the two variables 1) is negative by) is positive, ©) should have an absolute value greater than 0A, d) should be close to zero. Question 31. Let the coefficient of correlation of two variables X and Y be larger than zero, What will be the effect on it, if the data of X are multiplied by the factor of 2? 18) The effect depends on the data of X b) It depends on Y. ©) The coefficient will be doubled. «) It will be increased fourfold. 3.2. Probability Question 32. There are 8 socks in your drawer: 4 black and 4 red, You take 3 of them with you in the dark, Which statements are correct? a) It is sure that you get at least two socks (1 pair) of the same colour, ) It is sure that you get a pair of reds. ¢) The probability to get 3 of the same colour is 4) The probability to get 9 of the samo colour is, Question 33. In the sports injuries unit of a hospital, 40% of te paticuts are rugby players, 20% are swimmers and the remaining 40% play soccer. For a rugby player, the probability to be released on the first day is 10%; for a swinumer, it is 20%; for a soccer player, it is 80%. Which of the following statements are correct? 18) 40% of all patients aro releasod on the first di }) Given a patient is released om the first day, the probability of her/him being a soccer player is 80%. 10©) 80% of the non-swimmers have to stay for more than one day: Question 34. Let X be a random variable with probability density finetion, f(x) Which of the following statements are correct? 2) Phe expected value of X is 3 b) The probability of X < Lis 2. ©) The probability of X € [0,0.5] s d) The probability of X = 1 is zero. 3.3 Inference and Linear Models Question 35. Let X he a random variable defined by the density fimetion 8" ees fle) = {= eee 0 yee swith parameters a > 0 and 6 > 0, We observe a sample {3,4,8}. Which of the following statements are correct? a) The expected value of X exists for all combinations of a and 9. ) The expected value does only depend on a, but not on 8. ©) Given the sample, 8 cannot be larger than 3. 4) If we ane =2, the entnate of derived by the method of monnent given the mame x & 2 is also the maximum likelihood estimate of e in this ease 3 Question 36. We are interested in significant differences (level « = 0,05) between the expeeted values ly and pty of tro populations, Which of the following statements on statistical tests nro correct?” 8) We will formulate the null hypothesis as yay = po b) A test can always be applied in this situation, ©) A pevalue is the probability that the null hypothesis is correet, given the observed data 1d) If we obtain a pevalue of 0.04, we will reject (level a = 0.05) the null hypothesis, Question 37. One of the lines in the following scatter plot is the regression line fitted to the data, Which of the statements are corect? u18) The red and grecn line have the right direction, and, het 1c, one of them could be the rgression line, ) The bine line secms to represent the mean value of the data with respect to y and thus could be the regression line, ©) The point in the top right comer has a strong influence on the regression line. «) Leaving aside the point in the corner, the red line seems to fit better to the rest of the data Question 38. You have performed a linear regression analysis to explore sunflowers growth (in meters per month) depending on the watering (in litres per day). You have estimated the regression coeflicient to be 8 = 1.6. Wh ean you conclude? a) There is a significant correlation between watering and growth, b) An average sunflower growths 1.6 meters per month, ©) If you give it am additional litre of water per day, there will be an additional average growth of 1.6 meters per month. d) According to the model assumptions, an additional litre of water per day will result in additional 19.2 meters of growth after one year ©) You should consider further influencing quantities. 3.4 (Alternative 1) R Programming In this section, you ean opt between R (here) and Python (below) representations of the same qu Question 39, Which of the following R commands evahiates to TRUE? an) 56 ) TRUE & FALSE | FALSE & TRUE ©) FALSE & FALSE & FALSE | TRUE ) 1(CCTRUE > FALSE) > TRUE) & ITRUE) Question 40. Consider the following code chunk: 12roo while(x < 4) { x < sample(1:3, 1) print (x) + It is not a good idea to rin these lines hecanse a) xis an invalid argument to print () b) the condition x < 4 is never violated, ©) the function sample() docs not exist. 4) x is initialisod with the wrong type. Question 41. Which of the following code lines return TRUE? a) max(c(2, 3, 4, MA, 1, 5)) ma b) max(c(2, 3, 4, NA, 1, 5), [Link] = TRUE) ©) typeof (sum(c(1, 2, 3, 4, UA))) == "double" ) typeof (sum(1:4)) == "integer" ©) typeof (sum(e(1L, 2L, 3b, 4L, NA_real_), [Link] = TRUE)) "integer" Question 42. Which functions may have been used to generate the following plot and its under ®) 190 b) posnts() 6) ablineO 1) Antograted) Question 43. Considor the following code chunk and output and not that HA appears in the output of 10) Xi < rnorm(1e2) KD ML +3 Y < XL + x2 + rnorm(te2) am(y ~ Xt + x2)call an(formula = Y ~ x1 + X2) Coefficients: (intercept) x x2 2.979 2.019 A Which of the following statements are correct? 1a) Perfectly correlated regressons X41 and X2 are used, ) AQ) excludes X2 from the regression so that there is 1 least squares solution ©) NA indicates that the model fit to the data is perfect. 3.4 (Alternative 2) Python Programming In this section, you ean opt between R (above) and Python (here) representations of the same questions Question 39. Which of the following Python commands evaluates to True? a) 625 b) True & False | False & True ©) False & False & False | True dl) not(((True > False) > True) & (not (True))) Question 40. Consider the following code chunk: import randon x=0 while <4) x = [Link]([1, 2, 3]) print (x) It is not a good idea to run these lines because. 1) 2c an invalid argument to print) bb) the condition x < 4 is never violated. ©) the function [Link]() dovs not exist. 4) xis initialised with the wrong type. Question 41. Which of the following codetines return True. a) numpy -argnax (numpy array ((2,3,4,[Link],1,5])) ) numpy nanargnax (numpy array ([2,3,4,[Link],1,5)) 5 ©) type([Link]((1,2,3,4,[Link]]).sum()) is [Link] ) type (numpy array([1,2,3,4] ,étypesobject).sum()) is int ©) type(numpy-array((1,2,3,4]).sum() is int Question 42. Which packages may have been used to generate the following plot and its underlying data?0 2 4 6 x «) munpy b) matplotisb ©) statenodels 4) math Question 43. Consider the following code chunk and output and note that there are two warnings import numpy as mp import pandas as pd import [Link] as em Xi = [Link](0, 1, 100) x2=X1+3 Y = X1 + 42 + [Link](0, 1, 100) df = [Link]({"¥": ¥, "Ki": Ki, "2": x2) Linmodel = [Link](formula = "Y ~ XL + X2", data = af).£1t() Linnodel.. summary () Output Intercept -0.0272 x 1.1074 x2 1.0257 Warnings: [1] Standard Errors assumo that the covariance matrix of the errors is correctly specifica. [2] The smallest eigenvalue is 3.27e-31, This might indicate that there are strong multicollinearity problens or that the design aatrix is singular, Which of the following statements are correct? 1) Perfectly correlated regressions X1 and X2 are used. b) Either X1 oF X2 should be excluded, as the second regressor docs not ndd any information to the model. ©) Tho second warning indicates that the model fit to the data is perfect 154 Data Science Question 44. Consider a data sot data containing all German inhabitants, which is subsetted in the following, Process: data = subset (data, Gender == "fonale") data = subset (data, Status == "aarried*) data subset (data, Haircolor == "broun*) Soloct which stater wents aro true after all the subsets have appliod. 18) The data set contains all brown haired and married females worldwide. b) The data set contains all brown haired German inhabitants, ©) The data set contains all brown haire 1d married female German inhabitants ) The data set contains all brown haired, snarried, female German inhabitants with at least 2 children Question 45. For what ultimate purposes may algorithins like Nelder-Mead, Newton-Raphson or grudient-escent be used for? a) To find the minimum of a function, ) To find all zeros of a function, ° 4d) To solve a generalised regression problem, [o evaluate the derivative of a function, Question 46. ‘The Titanie data set contains information, whether passengers of the Titanie survived the shipwreck, based on their gender, age and passenger elass. The following decision tree has heen learned on this data. Which of the statements are true’? sexe male and O73 a90>0 95. pass «ar a poss = ra age 18 1) Overall, 62% of the passengers in the data sot ded, 1) A ‘new! passenger (Female, 3ed elias, 30-years old) is predicted to die in the shipwreck ©) 62% of the passengers in the data set are female ) All male 3a class passengers in the data set died, 16Question 47. Random forests are one of the most famous machine learning methods. They are easy to understand, easy to implement and reach good prediction performances even withont a hyper-parameter timing. Which of the following statements on random forest aro correct? 1) The prediction of a chasification forest is made by a majority vote of the trees’ predictions ») Tho prodiction of a regrossion forest is the median of the tree predictions ©) Bach single troe in the forest uses only a part of the data available 1) The training time of a random forest scales linear with the mumber of trees used. Question 48. Let us return to the Titanic data set. We now have learned several models and want to choose the best one. We used tce diferent methods to validate these models; The training error rate (apparent error rate), ‘the error rate on an external test set and the error rate estimated by a 10-fold cross validation, Learner Error on the test set | Cross Validation Error Decision Troe Random Forest (carest-Neighbour Which of the following statements are correct? 1 it should be used a) L-Nearest-Neighbour has a perfect training error and b) Random Forests outperforms both I-Nearest-Neighbour and the Decision Tree in terms of prediction error. ©) Not just in this ease, but in general, Cross Validation is the better validation strategy and should always be preferred aver the error on a single test sot d) Not just in this ease, but in general, Decision Troes always perform worse than Random Forests. Question 49. We try « last model class to find the pesfeet model for the Ttanie datu-set: An SVM. ‘The SVM is a model class that is very sensitive to hyper-parametcr tuning. Especially, the eost parameter C and the bandwidth of the RB kernel must he optinally adjusted in order to obtain a sensible model We use a nested resampling strateay to perform this hyper-puramcter tuning: At fist, 33% of the data are laid aside ns an external test set, to validate the result of the hyper-parameter ti AF (the outer resampling strategy). We use a random search as the tuning algorithm with a budget of 100 parameter spaces, ‘we nse all positive real mumbers for both C and 2. The performance of a single hyper-parameter setting is evaluated using a 10-fold cross validation (tho inner resumpling strategy). Moreover, in order to speed up the entire tuning proees, we utilise parallel computing, Which of the following statements are correct? 1) Using a nested resampling is necessary in order to detoct underfitting. b) As both C and \ are numeric parameters, any other optimization algorithm could be used instead of random search, ogy is arbitrary, and a bootstrapping would lead ©) Tho choice of cross-validation as the inner resampling stra to similar results. jon of the inner cross-validation «d) ‘Phe parallelization should take place at the innermost loop, henee, the exe loop should be parallelized, Question 50. Take a look at the following seatter plot of the so-called XOR dataset: 7It is a classifiention data-set with the goal of separating the red and the black observ inmber of rod and black observations is approximatcly equal. Which of the following fons. Assume, that the ements is correct’? a) A Decision Tree can reach a prediction error of (nearly) zero on this data-set b) When performing a variable sole 11,22 will be added to the model. ion using the step-wise forward selection algorithm, neither of the variables ©) A Linear Diseriminant Analysis (LDA) can reach a prediction extor of (nearly) zero on this data-set 4) Exery mode! using only one of the two variables 27,19 will have a missclassification error of approximately 50% We would like to thank you for taking your time and working through the test until the end. We hope that it, helped you to got in insight into the topics of our Bachelor program, and te get an idea about the advanced methods taught in our Master program. If you want to check your answers, and to understand the solutions, please have a Jook into the solution-pdf under this link: https: //#wy. statistik. [Link]/fileadnin/user_upload/ ‘Studiun/Studi ongaongo~Infos/Solf_Test_Master_Data Science Solutions. pdf