Socrative

We observe that average traffic in higher on weekdays than on weekends. We would like to back this up by calculating p-value What is NULL hypothesis?

There is no difference between weekend and weekday traffic

Let the observed difference D = 2.5, We have run 10000 permutation test with the following distribution. 1000 tests showed Dnull larger than 3 and 2000 tests showed Dnull >2. Estimate the p value.

p-value is between 0.1 and 0.2

We run 1000 permutation tests and 950 of these tests show mean(Score, AskQuestions='Often' > mean(Score, AskQuestions='Never) + 5 Alternative Hypothesis = "If you ask questions often your score is higher" What can we say about p-value?

p<0.05

We run permutation test 10,000, what is the smallest p-value we can obtain?

p=0.00001

Transformations of attributes which rpart cannot handle

Substrings - like area code of a phone number Page

What is false positive of test

Test is positive when it should be negative

First 5 rows of the class are sitting ONLY cs-stat students. Next two rows are sitting only economics students, next two rows only psychology students and finally the last row again in sitting ONLY cs students What is the error rate of best predictor of major of random student selected from the class audience?

0%

We are testing a hypothesis that Rutgers graduates make more than Princeton graduates after 10 years from graduation, Our data shows Rutgers graduates make D=$6500 more annually than Princeton graduates. In disbelief (what a deal, so much less in tuition and debt, and more in earnings!) we decide to run permutation test. (We run permutation test 10000 times, and in 250 cases we get D > 7000 and in 500 cases, D > 6000. What can you say about estimated p-value of hypothes that Rutgers graduates make more than Princeton graduates after 10 years from graduation?

0.025<p<0.05

Police investigates the crime which was committed on isolated island which has 10,000 inhabitants and no tourists. Their DNA matching techniques are old fashioned and there is 1:100 chance that two individuals match the same DNA. One of the suspect's DNA matches DNA found on crime scene, what is the probability that there may be someone else who also matches the found DNA?

1--(0.99)^9999

If medial test is always positive what is the probability of false positive

1-prior probability

If probability of a quiz is 0.2, what are the odds?

1:4

50% of students in data 101 class are cs majors. what are the odds of a student sitting in first row to be cs major? Assume that 10% of al cs majors sit in first row, while 5% of non-cs majors sit in the first row.

2:1

If odds of an event are 3:5, what is the probability of this event?

3/8

Assume that 51% of students in this class are cs and statistics, 24% are from psychology and 25% from economics On the basis of this information what is the ERROR of best predictor of major of randomly selected student?

49%

If prior odds are 1:10 and posterior odds are 1:2, what is likelihood ratio?

5

Right now, the distribution of majors in the first row of the 254 classroom is as follows 75% cs and statistics, 20% economics and 5 % psychology What is the precision of best predictior of major of randomly selected student from first row?

75%

The professor moody grade table has 6 attributes in addition to the grade. There are 1000 tuples in the table. What is the maximum depth of the decision tree assuming minimu bucket size is 5

995

We got negative z, z=-4 What does it mean?

Alternative hypothesis needs to be reversed (i.e instead A<B, B<A)

10% of music videos have more than one million views. The remaining music videos each have at most 1000 views each 80% of educational videos have more than 1000 views but less than 10,000 views. The remaining 20% of educational videos have less than 1000 views Which can we say about this?

Average number of views of music videos > Average number of views of educational video AND Random education videos has more views than random music videos

Ann voted for Trump. What are odds that she is evangelical christian? What is a Belief B, and what is observation A in this case?

B= Ann is Evangelical Christian A= Ann voted for Trump

What plot do you use for frequency distribution of CAT variable

Bargraph

which plot do you use to show score (NUM) distribution for students depending on how often they ask questions (CAT)

Boxplot

Can p value be very low (say 0.0000001) and we still cannot accept the alternative hypothesis?

Can happen if observation sample is biased

Which model would you select based on crossvalidation results?

Combination of high accuracy on all folds and low variance - high stability

We have run permutation test 10000 times. We got 500 tests with Dnull larger than 1. What observed value of D would result in pvalue less than 5%?

Dnull > 1

MSE is used for

Error calculation for numerical variable prediction

What is law of small numbers

Extreme results occur more often for small samples

Let traffic in Holland and Lincoln tunnels in our sample be distributed according to Holland --- rnorm(50, mean=60, sd=10) Lincoln - rnorm(50, mean=65, sd = 20) Alternative Hypothesis = Traffic in Lincoln tunnel is higher than traffic in Holland tunnel If mean of Holland tunnel sample goes up, say from 60 -> 63, the p-value would

Go up

Gubernatorial elections may occur at dfferent years in different states. Not like presidential elections. Where can Simpson paradox take place in

Gubernatorial elections

A model which has high accuracy and high stability in multiple crossvalidations will have

Hard to say

in Tillet 254 there are two attributes: ROW NUMBER and POSITION with values LEFT segment, CENTRAL segment of the class and RIGHT segment of the class Which of the location attributes (Row Number or Position) would be better "question" to ask in decision tree to predict major of random student assuming that: Distribution of students in each row number and in each location is the same as overall distribution of students in class 51% cs-stat, 24% economics 25% psychology

IRRELEVANT do not look at none of the two

What is Bonferroni coefficient?

It is used to correct for multiple hypothesis. It is not enough for p value to be less than 5% anymore, it has to be smaller than 5% divided by Bonferonni coeffcient

John is gun owner who voted for Hillary Clinton, what is more likely: John is a nerd John is liberal nerd John is conservative nerd John is hell of a confused nerd

John is a nerd

What is MSE?

Mean Square Error

what plot do you use to display frequency of a pair of CAT variables

Mosaic

So we have come up with best prediction rules for today's c;lass. Now, would these rules hold for the next class on Tuesday?

No, students who are late have to select a sit which is "first available" and not all students from Friday class, show up next Tuesday

What is Prior?

Probability of a belief

The goal of a permutation test is to:

Show how often the observed results could happen by random chance.

What can lead to larger decision tree?

Smaller value of minsplit parameter

which one is not ML method? Random Forest SVM Smart Bayes Naive Bayes Linear Regression

Smart Bayes

We observe that average income of immigrants is higher than average income of non-immigrants. We want to validate this observation by calculating p-value. What is null hypothesis

There is no difference in average income between immigrants and non-immigrants

In our data set for 2018 data 101 class results we clearly saw that Students with GPA >3.0 get higher score in data 101 than students with GPA <=3.0. We want to back this observation up - by calculating p-value. What is NULL hypothesis?

There is no difference in score of high GPA and low GPA students

HYPOTHESIS: There is more cs-stat students in this class than NON cs-stat students What is NULL HYPOTHESIS?

There is same number of cs-stat students in this class than NON cs-stat students

Two decision trees are the same except minsplit parameter We run the same command rpart with two different values of minsplit: misplit1< minsplit 2 and all other parameters are the same

Tree 2 will be at least as large as Tree 1

Simpson Paradox occurs when

Trend appears in several different groups of data but disappears or reverses when these groups are combined. OR Ratios are compared over groups of different sizes

We observe that Texting in class matters for your final score. Which z-test should be aplied to evaluate the hypothesis that our observation is true in general.

Two-sided test

If p value is equal to 1, what does it mean?

We fail to reject null hypothesis

Our hypothesis that average life expectancy is higher for rich countries than for poor countries will be rejected if

We fail to reject null hypothesis that " average life expectancy is same for rich countries and for poor countries"

If p-value =0

We reject null hypothesis

We got p value of 0.0001 and showed that French people are on average more happy than British. There are 200 nations represented in HAPPINESS TABLE

We should count the number of all possible hypotheses (pairs of countries) and apply Bonferroni coefficient

Can one beat rpart's accuracy using freestyle prediction?

When none of the attributes is predictive - and new attributes have to be constructed. rpart will not invent new attributes.

Can non-cs majors have higher average prediction accuracy in a prediction challenge than cs-najors but random cs major have higher accuracy in prediction challenge than random non-cs

Yes

Deep learning conference accepts only 5% of submitted paper Data Science conference accepts 10% submitted papers while Big Data conference accepts 50% of submitted papers Is it possible that Rutgers has higher chance of getting paper accepted in any of these three conferences than Princeton does even though Princeton has higher acceptance for each of the three conferences than Rutgers?

Yes

Now lets assume that first row in central location has 90% of cs-stat students and 10% of psychology and 10% of economics students, but last two rows have only 35% of cs-stat students Can we improve prediction error using ROW and POSITION now?

Yes, the tree which has both POSITION AND ROW attributes would lead tp LOWEST prediction error

which plot do you use to show grade (CAT) distribution for students who text frequently (CAT)

bargraph

What plot do you use to check if Mr Moody assigns grades (CAT) solely on the basis of score (NUM) in class?

boxplot

When z-value goes up, p value goes

down

True or False, every distribution converges to normal (Bell curve) when the number of observations growth is to infinity

false

moody$PF<-"" moody[moody$GRADE!='F',]$PF <- 'pass' moody[moody$GRADE==F,]$PF <- 'fail' true or false

false

How to calculate standard deviation

get mean of values subtract mean from each value and square the result get mean of new values and square the result

Let traffic in Holland and Lincoln tunnels in our sample be distributed according to Holland --- rnorm(50, mean=60, sd=10) Lincoln - rnorm(50, mean=65, sd = 20) Alternative Hypothesis = Traffic in Lincoln tunnel is higher than traffic in Holland tunnel If sd of Lincoln tunnel sample goes up, say from 20 -> 40, the p-value would

go up

what does this code do: ask_question_grade <- tapply(moody$SCORE,moody$ASKS_QUESTIONS,max)

maximum score a student got for each of the values of ASK QUESTIONS attributes

Tony has scored higher than Anand in each grading categories (tests, assignments, particpation) in data 101. Can Simpson paradox occur here? i.e Anand can still have higher total score than Tony?

no

among 10,000 permutations - 100 return D > 10 and 500 return D> 5 If Dnull= 7 what are p-value bounds?

p <0.05 and p>0.01

We run permutation test 10,000 times. Observed difference of means is 8.5. We see that there are 1000 results with observed difference D >5 and 200 permutation results with observed difference D > 10. What can we conclude about p-value?

p> 0.02 and p<0.1

If medical test is always positive - what is the probability of true positive

prior

what is p-value

probability that observed result can be obtained under condition that null hypothesis is true

What is a difference before rpart function (or any other ML method) and predict function?

rpart can only work on training data. predict() applies rpart tree model to any data set

What is central limit theorem

sample means distribution

which R function would you use to plot average score (NUM) for each grade (CAT) in professor Moody class?

tapply

what is the relation between z-value and p-value

the larger z is the smaller p is

Permutation test function requires one variable to be numerical and another categorical

true

m<-data.frame(first<-c('A', 'B'), second <-c(1:10) )

true

what do you need when you use z-test for difference means of two populations

we need standard deviation of each of the two populations

3P percentages in season 2018 2019 Joe Harris from Booklyn Nets 41,9% 47.4% Steph Curry from Warriors 41.6% 43.7% Can Steph Curry have higher overall percentage of three pointers (3P) for both seasons?

yes