CSCI 343 Quiz 1

Question 1

continuous data types

Answer 1

float, double

Question 2

discrete data types

Answer 2

int

Question 3

categorical data types

Answer 3

specific set, enum (ex: red, orange, yellow; classified, unclassified, part-time)

Question 4

binary data types

Answer 4

boolean, 0/1, T/F, logical

Question 5

ordinal data types

Answer 5

categorical with order (ex: rating 1, 2, 3, 4, 5; fresh, soph, jr, sr)

Question 6

mean (aka average)

Answer 6

sum / count

Question 7

trimmed mean

Answer 7

drop a few of the high values and the low values and calculate the mean of the remaining (ex: Olympics)

Question 8

weighted mean (aka weighted average)

Answer 8

sum of all values times corresponding weights divided by sum of weights (ex: calculating grades)

Question 9

median

Answer 9

middle number (when sorted), if an odd # elements
if even # elements, median is the average of the two middle elements

Question 10

(median/mean) is better for skewed data sets b/c ?

Answer 10

median b/c it won’t include outliers

Question 11

deviations

Answer 11

difference between the observed values and the median

Question 12

variance

Answer 12

sum of the squared deviations from the mean, divided by n-1

Question 13

standard deviation

Answer 13

square root of the variance

Question 14

range

Answer 14

max - min

Question 15

percentile

Answer 15

the pth percentile is a value (not necessarily in the data set) such that at least p% of the data items are of this value or less and (100-p)% of the data items are this value or more

Question 16

quartiles are

Answer 16

25% pieces

Question 17

mode

Answer 17

the data item that occurs the most

Question 18

box plots show

Answer 18

the four quartiles, the median, and the box is the inner quartile range

Question 19

machine learning

Answer 19

learning from experience/history

Question 20

supervised learning

Answer 20

prediction, regression (ex: professor tells you answers); uses labels

Question 21

the vast majority of data science work is in

Answer 21

supervised learning

Question 22

unsupervised learning

Answer 22

clustering; does not use lables

Question 23

reinforcement learning

Answer 23

robots, AI; no answer, but a reward (like closeness to victory, hot/cold)

Question 24

basic formulation of machine learning

Answer 24

• assume complete, correct data
• correct label, unique label
• prediction (or regression)
• non-mixed types of attributes

Question 25

goal of machine learning is to

Answer 25

generalize from some samples so we can predict label on unseen examples

Question 26

process of machine learning

Answer 26

build a model, train our model, and test the model built

Question 27

when we train a model, we

Answer 27

put in samples with labels, run a learning algorithm, and then create a classifier/predictor

Question 28

when we test a model, we

Answer 28

put in new examples to the classifier/predictor and retrieve the prediction of the label for the new example (then compare in confusion matrix)

Question 29

accuracy

Answer 29

amount correct / total runs

Question 30

confusion matrix

Answer 30

compare predicted to actual, want numbers along the diagonal to be big; can also help you evaluate your model

Question 31

k-fold cross validation

Answer 31

for i=1 to k
build model on all data except subset i
predict on subset i

average results

Question 32

LOOCV

Answer 32

leave one out cross validation (k=n)

Question 33

noise

Answer 33

data that is not correct

Question 34

impute

Answer 34

put in an estimate or meaningless data to show it’s blank

Question 35

curse of dimensionality

Answer 35

too much data for each example

Question 36

how to help with the curse of dimensionality

Answer 36

select a helpful subset through dimension reduction or feature extraction

Question 37

overfitting

Answer 37

doing well on training data, but poorly on the test data (work too hard, not general enough)

Question 38

must do better than baseline

Answer 38

tempting to predict the most common label; improve upon the baseline (ex: flipping a coin 50% of the time you’d be right, but if you could increase it to 53% that’s better)

Question 39

interpretability

Answer 39

should be able to understand the model, not say the algorithm told me so but idk why