Name:
Andrew ID:
Collaborated with:
This lab is to be done in class (completed outside of class if need be). You can collaborate with your classmates, but you must identify their names above, and you must submit your own lab as an knitted HTML file on Canvas, by Thursday 10pm, this week.
This week’s agenda: basic indexing, with a focus on matrices; some more basic plotting; vectorization; using for() loops.
We’re going to look at a data set on 97 men who have prostate cancer (from the book The Elements of Statistical Learning). There are 9 variables measured on these 97 men:
lpsa: log PSA scorelcavol: log cancer volumelweight: log prostate weightage: age of patientlbph: log of the amount of benign prostatic hyperplasiasvi: seminal vesicle invasionlcp: log of capsular penetrationgleason: Gleason scorepgg45: percent of Gleason scores 4 or 5To load this prostate cancer data set into your R session, and store it as a matrix pros.dat:
pros.dat =
as.matrix(read.table("http://www.stat.cmu.edu/~ryantibs/statcomp-S18/data/pros.dat"))1a. What are the dimensions of pros.dat (i.e., how many rows and how many columns)? Using integer indexing, print the first 6 rows and all columns; again using integer indexing, print the last 6 rows and all columns.
1b. Using the built-in R functions head() and tail() (i.e., do not use integer indexing), print the first 6 rows and all columns, and also the last 6 rows and all columns.
1c. Does the matrix pros.dat have names assigned to its rows and columns, and if so, what are they? Use rownames() and colnames() to find out. Note: these would have been automatically created by the read.table() function that we used above to read the data file into our R session. To see where read.table() would have gotten these names from, open up the data file: http://www.stat.cmu.edu/~ryantibs/statcomp-S18/data/pros.dat in your web browser. Only the column names here are actually informative.
1d. Using named indexing, pull out the two columns of pros.dat that measure the log cancer volume and the log cancer weight, and store the result as a matrix pros.dat.sub. (Recall the explanation of variables at the top of this lab.) Check that its dimensions make sense to you, and that its first 6 rows are what you’d expect. Did R automatically assign column names to pros.dat.sub?
1e. Using the log cancer weights and log cancer volumes, calculate the log cancer density for the 97 men in the data set (note: by density here we mean weight divided by volume). There are in fact two different ways to do this; the first uses three function calls and one arithmetic operation; the second just uses one arithmetic operation. Note: in either case, you should be able to perform this computation for all 97 men with a single line of code, taking advantage of R’s ability to vectorize. Write code to do it both ways, and show that both ways lead to the same answer, using all.equal().
1f. Append the log cancer density to the columns of pros.dat, using cbind(). The new pros.dat matrix should now have 10 columns. Set the last column name to be ldens. Print its first 6 rows, to check that you’ve done all this right.
2a. Using hist(), produce a histogram of the log cancer volume measurements of the 97 men in the data set; also produce a histogram of the log cancer weight. In each case, use breaks=20 as an arugment to hist(). Comment just briefly on the distributions you see. Then, using plot(), produce a scatterplot of the log cancer volume (y-axis) versus the log cancer weight (x-axis). Do you see any kind of relationship? Would you expect to? Challenge: how would you measure the strength of this relationship formally? Note that there is certainly more than one way to do so.
2b. Produce scatterplots of log cancer weight versus age, and log cancer volume versus age. Do you see relationships here between the age of a patient and the volume/weight of his cancer?
2c. Produce a histogram of the log cancer density, and a scatterplot of the log cancer density versus age. Comment on any similarities/differences you see between these plots, and the corresponding ones you produced above for log cancer volume/weight.
2d. Delete the last column, corresponding to the log cancer density, from the pros.dat matrix, using negative integer indexing.
3a. The svi variable in the pros.dat matrix is binary: 1 if the patient had a condition called “seminal vesicle invasion” or SVI, and 0 otherwise. SVI (which means, roughly speaking, that the cancer invaded into the muscular wall of the seminal vesicle) is bad: if it occurs, then it is believed the prognosis for the patient is poorer, and even once/if recovered, the patient is more likely to have prostate cancer return in the future. Compute a Boolean vector called has.svi, of length 97, that has a TRUE element if a row (patient) in pros.dat has SVI, and FALSE otherwise. Then using sum(), figure out how many patients have SVI.
3b. Extract the rows of pros.dat that correspond to patients with SVI, and the rows that correspond to patients without it. Call the resulting matrices pros.dat.svi and pros.dat.no.svi, respectively. You can do this in two ways: using the has.svi Boolean vector created above, or using on-the-fly Boolean indexing, it’s up to you. Check that the dimensions of pros.dat.svi and pros.dat.no.svi make sense to you.
3c. Using the two matrices pros.dat.svi and pros.dat.no.svi that you created above, compute the means of each variable in our data set for patients with SVI, and for patients without it. Store the resulting means into vectors called pros.dat.svi.avg and pros.dat.no.svi.avg, respectively. Hint: for each matrix, you can compute the means with a single call to a built-in R function. What variables appear to have different means between the two groups?
3d. Consider the 9 “magic denominators” given below. For each of the 9 variables in our data set, compute the difference between its mean in SVI patients and non-SVI patients, divided by the corresponding denominator. For example, for the first variable, lcavol or log cancer volume, this is going to be: \[
\frac{\text{lcavol avg for SVI patients} - \text{lcavol avg for non-SVI patients}}
{\text{denominator #1}},
\] and so on. Store the result of this computation as a vector called pros.dat.t.stat. Note: you should be able to perform this computation for all 9 variables with a single line of code, taking advantage of R’s ability to vectorize. Print the results. For which variables is the absolute value of pros.dat.t.stat larger than 2? Again, this should take just a single line of code. Challenge: what do you think is being computed here (look at the name of the variable you created!), and what is the significance of the value 2?
magic.denom = c(0.19092077, 0.08803179, 1.91148819, 0.34076326, 0.00000000,
0.25730390, 0.15441770, 6.30903678, 0.23021447)i and sets it equal to 1. The second line defines a string var.name and sets it equal to a placeholder value “FOO”. Edit this line so that var.name is set to the column name of the ith variable in the pros.dat matrix. The third line defines a string title based on var.name using the paste() function, which pastes together two strings (we’ll see much more on paste() and related functions soon). Write a fourth line which plots a histogram of the ith column in pros.dat, and passes the additional arguments: breaks=20, main=title (to set the title), and xlab=var.name (to set the x-axis label), to the hist() function. Try running this block of 4 lines with multiple different settings of i (i.e., change i=1 to i=2, and so on), and check that the output makes sense to you.i = 1
var.name = "FOO"
title = paste("Histogram of", var.name)4b. Write a for() loop to produce a histogram of each column in the pros.dat matrix. If it helps, think about doing this in two steps: as the first step, write a for() loop that iterates an index variable i over the integers between 1 and the number of columns of pros.dat (don’t just manually write 9 here, pull out the number of columns programmatically), with an empty body. As the second step, paste your solution code for the last question (lines 2 through 4) into the body of the for() loop. Once run, the loop should produce 9 histograms for you.
4c. Produce a scatterplot of the log cancer volume versus SVI. Since SVI is binary, you’ll notice that we just see two vertical strips of points: one at \(x=0\), and the other at \(x=1\). R has a special data type that we haven’t learned yet, called a “factor”, which is specifically designed to handle categorical data. We can (try to) cast any object to a factor using as.factor(). Produce a scatterplot of the log cancer volume versus SVI again, but this time, cast the SVI variable to a factor when it is passed to plot(). You should notice quite a difference in the result: what is being shown now is called a “boxplot”.
4d. Similar to what you did in Q2a: define i=1, define var.name to be the the column name of the ith variable in the pros.dat matrix, and define title by pasting together var.name and “versus SVI” with the paste() function. Then, plot the ith column in pros.dat versus SVI, with the SVI variable being converted to a factor. In the call to plot(), pass the additional arguments: main=title, xlab="SVI", and ylab=var.name. Try running this block of code with multiple different settings of i, and check that the output makes sense to you.
4e. Write a for() loop to produce boxplots of the columns in pros.dat versus SVI, but only for variables in which there is a significant difference between SVI and non-SVI patients. Specifically, we will consider the difference for a variable to be significant if its entry in pros.dat.t.stat, as you computed in Q3d, is larger than 2 in absolute value. If it helps, consider breaking down this task into three steps: first, write a for() loop that iterates an index variable i over the integers between 1 and the number of columns of pros.dat. Second, write in the body an if() statement that checks whether the ith entry of pros.dat.t.stat is larger than 2 in absolute value. Third, paste your solution code for the last question (lines 2 through 4) into the body of the if() statement. Once run, the loop should produce 6 boxplots.
4f. One of the plots you produced in the last question is kind of useless. Specifically, there is a plot of SVI versus itself, which is not useful. Modify your solution code for the last question so that this plot is excluded. Hint: you can do this by adding to your if() statement appropriately, using a Boolean operator. Once run, your loop should produce 5 boxplots. Visually, which variable appears to have the biggest difference between SVI and non-SVI patients? Does this agree with the variable that has the largest absolute value in pros.dat.t.stat (apart from SVI, whose value is vacuously infinite)?
Challenge. Use the last code block from this week’s lecture as starter code to complete the following task. In the body of a repeat loop, prompt the user for a variable name to plot, using readline(). Check if the string that you collect from the user is one of the column names in pros.dat. Hint: use the %in% operator. If the string is indeed one of the column names, produce a histogram of the corresponding variable, with a title and x-axis label set appropriately. If the string is “quit”, then break out of the repeat loop. Otherwise, print to the console: “Oops! That’s not a variable in my data set.” In the Rmd code chunk for your solution code, make sure to set eval=FALSE, as was done in the Rmd file for the lecture notes (see the last code chunk). Otherwise your lab file will never finish knitting. Try out your solution code by running it in your console.
Challenge. Extend your prompting code in the last question to allow for scatterplots as well as histograms, and any other options you deem interesting, that the user might want to specify.