TidyverseSkeptic

by matloff

An opinionated view of the Tidyverse "dialect" of the R language.

253 Stars 22 Forks Last release: Not found 93 Commits 0 Releases

Available items

No Items, yet!

The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:

TidyverseSkeptic

An alternate view of the Tidyverse "dialect" of the R language, and its promotion by RStudio.

Norm Matloff, Prof. of Computer Science, UC Davis (former Prof. of Statistics at UCD)

NOTE: This Is the Short Form

To me, the main problem with the Tidyverse is teachability. I believe that teaching beginners with no coding backgroun is actually impeded by the use of the Tidyverse instead of base-R. This short form here consists of extraction of the teachability aspects of my TidyverseSkeptic essay. The full document, addressing other issues, is available here.

Disclaimer

This essay is somewhat frank, involving the very popular Tidyverse and RStudio. I hope it is polite and taken as constructive criticism.

I like and admire the RStudio people, including the Tidyverse originator, Hadley Wickham, and have always supported them, both privately and publicly. They and I have been interacting from the beginning, when the firm consisted of only founder JJ Allaire and ace developer Joe Cheng. I highly praise the firm to my students, and I use and recommend Hadley's (non-Tidyverse) packages ggplot2 and stringr, and on occasion devtools has been an absolute lifesaver for me.

In other words, I don't consider RStudio to be some evil cabal. I state at various places in this essay that I think their actions have been well-intentioned. Nevertheless, I believe that RStudio took a wrong turn when it decided to promote the Tidyverse, which has led to a situation in which the unity and health of the language is at stake.

My bio is here. Specifically in terms of R, I've been an R user and developer since near the beginning, having used R's predecessor S before that. I've published several books that use R, and have served as the Editor-in-Chief of the R Journal. (Hadley is a former EiC on the journal.)

TEACHABILITY OVERVIEW

  • My biggest concern is involves teaching of R. Tidy makes it more DIFFICULT for nonprogrammer learners of R to become proficient in the language.

  • The Tidyverse arose from the desire to have a set of functions/packages that are consistent with each other, a "purist" philosophy that appeals, for instance, to computer scientists. The Tidyverse also borrows from other "purist" computer science (CS) philosophies, notably functional programming (FP). The latter is abstract and theoretical, difficult even for CS students, and thus it is clear Tidy is an unwise approach for nonprogrammer students of R.

  • Another price of purity is increased complexity and abstraction, making code more prone to error (as well as a sacrifice in performance).

  • In fact, the Tidyverse is general is far too complex for learners without prior coding background, causing what psychologists call cognitive overload.

  • Indeed, even many Tidy advocates concede that it is in various senses often more difficult to write Tidy code than base-R.
    Hadley says, for instance, "it may take a while to wrap your head around [FP]."

  • Thus, contrary to the claim made by RStudio for promoting the Tidyverse is that it makes R easier to teach to non-programmers, I would argue that, on the contrary, the Tidyverse makes R harder to learn for this group.

  • RStudio has done an admirable job in bringing more women and underrepresented minorities into R. Yet by saddling them with the complicated Tidyverse system, RStudio has made it more difficult for these groups to make contributions to the language, in the form of writing CRAN packages, authoring books and so on, which require good facility with base-R even if the code is largely Tidyverse.

TEACHABILITY

Teaching has been a keen interest of mine since my college days. I've been a teacher of stat and computers for many years, and have won various teaching awards. My textbook, Statistical Regression and Classification: from Linear Models to Machine Learning, was the recipient of the 2017 Ziegel Award.

But it goes far beyond that; I really am intensely interested in how people learn, from children to middle-aged adults. Among other things, I've taught English As a Second Language to immigrant adults, most of whom have had less than a high school education.

In discussing teaching, I am taking the target audience here to be nonprogrammers who wish to use R for data analysis, not those who wish to become professional R programmers.

Case Study: Teaching One's First Lesson in R

The Tidyverse is just too complex for beginners. Here some quick examples of the complexity of Tidy, and its consequent unsuitability for use in teaching nonprogrammer learners of R.

Let's look at my online R tutorial, fasteR). Consider, for instance, an innocuous line like

> hist(Nile)

i.e. drawing a simple histogram of R's built-in Nile River dataset.

This is in the very first lesson in my tutorial. Easy! By contrast, the Tidy crowd forbids use of base-R plots, insisting on using ggplot2 (which again is not Tidy, but is considered as such by the Tidy advocates). To be Tidy the instructor would have to do something like

> library(ggplot2)
> dn  ggplot(dn) + geom_histogram(aes(Nile),dn)

Here the instructor would have a ton of things to explain -- packages, data frames, ggplot(), the aes argument, the role of the '+' (it's not addition here) and so on -- and thus she could definitely NOT present it in the first lesson.

Also in my very first lesson, I do

> mean(Nile[80:100])

printing the mean Nile River flow during a certain range of years. Incredibly, not only would this NOT be in a first lesson with Tidy, the students in a Tidy course may actually never learn how to do this. Typical Tidiers don't consider vectors very important for learners, let alone vector subscipts.

As a concrete example of this Tidy point of view, consider the book Getting Started with R, by Beckerman et al, Oxford University Press, second edition, 2017. The book makes a point of being "Tidyverse compliant". In 231 pages, vectors are mentioned just briefly, with no coverage of subscipts.

Case Study: Dalgaard book

A researcher tweeted in December 2019 that an introductory statistics book by Peter Dalgaard is "now obsolete," because it uses base-R rather than Tidy. Think of what an update to Tidy would involve, how much extra complexity it would impose on the students. Here is an example from the book:

> thue2 

This could easily be in the base-R instructor's second lesson, if not the first. For Tidy, though, this would have to be changed to

> library(dplyr)
> thue2 % filter(blood.glucose < 7)

Here the instructor would first have to teach the pipe operator '%>%', again extra complexity. And in so doing, she would probably emphasize the "left to right" flow of pipes, but the confused students would then wonder why, after that left-to-right flow, there is suddenly a right-to-left flow, with the '<-'. (For some reason, the Tidy people don't seem to use R's '->' op.)

Again, the Tidyverse is simply too complex for R learners without coding background. It slows down their learning process.

The Tidyverse advocates' claims

As a lifelong passionate teacher, I strongly question the claim made by Tidyverse advocates that it facilitates the teaching of R to non-programmers, as opposed to teaching them base-R.

(I regard both dplyr and data.table as advanced topics; neither is suitable for beginners. On the other hand, I think teaching beginners ggplot2 is fine, but point out again that it is not part of the Tidyverse.)

There has been no study of Tidy advocates' teachability claim. Advocates often provide testimonials from students like

  • "I learned R using Tidyverse, and now am productive in R"

  • "I like the English-like nature of the Tidyverse"

  • "I can make beautiful graphics with the Tidyverse"

  • "Tidyverse showed me the fluidity of data"

Advocates who are instructors will echo those statements, and add, e.g.

  • "Base-R is good for professional programmers, but the Tidyverse is the best R learning tool for non-techies who just want to do data analysis"

  • "R was created by statisticians, for statisticians. But we are data scientists, mainly interested in producing graphs and tables"

  • "The Tidyverse is modern R, for the rest of us"

All of these statements are either misleading, irrelevant to the teachability issue, or downright meaningless. They say nothing at all about the teachability of base-R in comparison to Tidy. (It is ironic that advocates who present such statements are data scientists, who ought to know the need for a control group.)

Tidyverse makes learning harder, due to adding much complexity

Contrary to the Tidy advocates' claim, I believe using the Tidyverse makes things more difficult for learners without prior programming background.

There is a serious problem of cognitive overload. Tidyverse students are being asked to learn a much larger volume of material, which is clearly bad pedagogy. See "The Tidyverse Curse", in which the author says inter alia that he uses "only" 60 Tidyverse functions -- 60! The "star" of the Tidyverse, dplyr, consists of 263 functions.

While a user initially need not use more than a small fraction of those functions, the high complexity is clear. Every time a user needs some variant of an operation, she must sift through those hundreds of functions for one suited to her current need.

Tidy advocates say the uniformity of interface in all those functions makes learning them easier. Uniform syntax is nice, yes, but the fact remains that users must learn the semantics of the functions, i.e. what operations they perform. What, for example, is the difference between summarize(), summarize_each(), summarize_at() and summarize_if()? Under which circumstances should each be used? The user must sift through this.

As Matt Dowle, creator of data.table, pointed out about dplyr,

It isn't one function mutate that you combine in a pipe. It's mutate, mutate_, mutate_all, mutate_at, mutate_each, mutateeach, mutate_if, transmute, transmute_, transmute_all, transmute_at and transmute_if. And you're telling me [because of consistency of the user interfaces] you don't need a manual to learn all those?

Having a common syntax thus does not compensate for this dizzying complexity.

By contrast, if the user knows base-R (not difficult), she can handle any situation with just a few simple operations. The old adage applies: "Give a man a fish, and he can eat for a day. Teach him how to fish, and he can eat for a lifetime."

What Tidy promoters want R beginners NOT to learn

The Tidyers make a point of avoiding the most basic parts of base-R:

  • the '$' operator

  • '[[ ]]'

  • loops

  • plot() and the associated basic graphics functions

They would argue that this "simplifies" the learning process, but actually it forces beginners to come up with more complex, less intuitive and harder-to-read solutions.

Case Study: the tapply() Function

One of the most commonly-used functions in R is tapply(). As noted below, for some reason Tidy advocates hate this function, but arguably it is perfect for R beginners.

Consider a common example in tutorials on the Tidyverse, involving R's mtcars dataset. The goal is to find mean miles per gallon, grouped by number of cylinders. The Tidy code offered is

mtcars %>%
   group_by(cyl) %>% 
   summarize(mean(mpg))

Here is the base-R version:

tapply(mtcars$mpg,mtcars$cyl,mean)

Both are simple. Both are easily grasped by beginners. After being presented with some examples, beginners have no trouble using them in a new setting of similar type. The Tidy version requires two function calls rather than one for base-R. But again, both versions are simple, so let's call it a tie. But is it certainly not the case that the Tidy version is easier to learn.

It's instructive to look at what happens when one groups by two aspects:

> mtcars %>%
+ group_by(cyl,gear) %>%
+ summarize(mean(mpg))
# A tibble: 8 x 3
# Groups:   cyl [3]
    cyl  gear `mean(mpg)`
          
1     4     3        21.5
2     4     4        26.9
3     4     5        28.2
4     6     3        19.8
5     6     4        19.8
6     6     5        19.7
7     8     3        15.0
8     8     5        15.4
> tapply(mtcars$mpg,list(mtcars$cyl,mtcars$gear),mean)
      3      4    5
4 21.50 26.925 28.2
6 19.75 19.750 19.7
8 15.05     NA 15.4

With tapply(), students do have to be told that in the case of more than one grouping variable, they need to surround the variables with 'list'. Again, once they are given examples, students have no trouble with this.

But look at the form of the output: The Tidy version outputs a tibble, rather hard to read, while tapply() outputs an R matrix, printed out as a two-way table. The latter form is exactly what many students need in their applications, e.g. social science research.

In searching through the hundreds of functions in dplyr, it is not clear to me which one, if any, can convert that Tidy output to the very informative tabular view that tapply() provides. If there is one, the fact that one is not easily identifiable illustrates my point above that Tidy is actually very bloated, not suitable for beginners.

Moreover, the tapply() output is more informative in a second sense, letting the user know that there were no 8-cyliner, 4-speed cars, again the kind of thing that is quite meaningful in many applications.

Actually, the Tidy version can be modified in order to notice that empty group:

> mtcars$cyl  mtcars$gear  mtcars %>% 
+    group_by(cyl,gear,.drop=FALSE) %>% 
+    summarize(mean(mpg))
# A tibble: 9 x 3
# Groups:   cyl [3]
  cyl   gear  `mean(mpg)`
          
1 4     3            21.5
2 4     4            26.9
3 4     5            28.2
4 6     3            19.8
5 6     4            19.8
6 6     5            19.7
7 8     3            15.0
8 8     4           NaN  
9 8     5            15.4

Note the need to convert to factors, something not mentioned in the Tidy documentation, and which would further complicate things for R beginners even if it were documented.

So, in terms of clarity and learnability, the Tidy and base-R versions in this paricular example are both good, again showing that Tidy is not easier to learn. And In terms of usability, I'd give base-R the win here.

Use of functional programming

Another featured Tidyverse package, the functional programming (FP)-oriented library purrr, has 177 functions. Again the point about complexity applies; we again have the "too many functions to learn" problem as we saw with dplyr above.

At the basic level, FP is merely replacing loops by calls to FP functions. R's apply family, plus Reduce(), Map() and Filter() should be considered FP. In many cases, using such functions is the right solution. But the indiscriminate use of FP, advocated by many Tidyers, to replace all loops is clearly overdoing it, and makes things especially difficult for beginners.

This is clear a priori -- FP involves writing functions, a skill that most beginners take a long time to develop well.

It is worth noting that top university Computer Science Departments have shifted away from teaching their introductory programming courses using the FP paradigm, in favor of the more traditional Python, as they deem FP to be more abstract and challenging.

An interesting discussion of the topic is in Charavarty and Keller. Though they support using FP in introductory programming classes for CS majors, the authors' goals are antithetical to those of R learners. The authors list three goals, one of which is to teach theoretical computer science, certainly not desirable for teaching R in general, let alone for teaching R to those with no coding experience. They also concede that a key concept in FP, recursion, is a "signficant obstacle" even for CS students.

If FP is tough for CS students, it makes no sense to have nonprogrammer learners of R use it.

Even Hadley, in R for Data Science, says:

The idea of passing a function to another function is extremely powerful idea, and it’s one of the behaviours that makes R a functional programming language. It might take you a while to wrap your head around the idea, but it’s worth the investment.

Actually, most non-FP languages allow passing one function to another, but yes it is a powerful tool, worth the investment of time -- for the experienced R programmer. But again, it's wrong to foce nonprogrammer learners of R to "wrap their heads around" purrr.

purrr vs. base-R example

Again, let's use an mtcars example taken from an online tutorial. Here the goal is to regress miles per gallon against weight, calculating R2 for each cylinder group. Here's the Tidy solution, from the online help page for map():

mtcars %>%
  split(.$cyl) %>%
  map(~ lm(mpg ~ wt, data = .)) %>%
  map(summary) %>%
  map_dbl("r.squared")

output

4 6 8 0.5086326 0.4645102 0.4229655

There are several major points to note here:

  • The R learner here must learn two different map functions for this particular example, and a dozen others for even basic use. This is an excellent example of Tidy's cognitive overload problem. Actually, purrr has 52 different map functions! (See below.)

  • The first '~' in that first map call is highly nonintuitive. Even experienced programmers would not be able to guess what it does. This is starkly counter to the Tidyers' claim that Tidy is more intuitive and English-like.

  • Tidy, in its obsession to avoid R's standard '$' symbol, is causing all kinds of chaos and confusion here.

    The hapless student would naturally ask, "Where does that expression 'summary' come from?" It would appear that map() is being called on a nonexistent variable, summary. In actuality, base-R's summary() function is being called on the previous computation behind the scenes. Again, highly nonintuitive, and NOT stated in the online help page.

    The poor student is further baffled by the call to map_dbl(). Where did that 'r.squared' come from? Again, Tidy is hiding the fact that summary() yields an S3 object with component r.squared. Yes, sometimes it is helpful to hide the details, but not if it confuses beginners.

The fact is, R beginners would be much better off writing a loop here, avoiding the conceptually more challenging FP. But even if the instructor believes the beginner must learn FP, the base-R version is far easier:

lmr2 

Here lmr2() is defined explicitly, as opposed to the Tidy version, with its inscrutable '~' within the map() call.

In a Twitter discussion of the above example, a Tidy advocate protested that the above purrr code was not appropriate for learners:

Sure, but my original tweet was about teaching newbies. Your example is not really relevant to that because it's about a VERY complex concept.

Exactly my point! Newbies should write this as a loop, NOT using purrr. But the Tidy promoters don't want learners to use loops. So the instructor using Tidy simply would avoid giving students such an example, whereas it would be easy for the base-R instructor to do so.

As noted, the Tidy version shown earlier is an illustration of the "too many functions to learn," cognitive overload, problem we saw earlier with dplyr. Behold:

> ls(package:purrr,pattern='map*')
 [1] "as_mapper"      "imap"           "imap_chr"       "imap_dbl"      
 [5] "imap_dfc"       "imap_dfr"       "imap_int"       "imap_lgl"      
 [9] "imap_raw"       "invoke_map"     "invoke_map_chr" "invoke_map_dbl"
[13] "invoke_map_df"  "invoke_map_dfc" "invoke_map_dfr" "invoke_map_int"
[17] "invoke_map_lgl" "invoke_map_raw" "lmap"           "lmap_at"       
[21] "lmap_if"        "map"            "map_at"         "map_call"      
[25] "map_chr"        "map_dbl"        "map_depth"      "map_df"        
[29] "map_dfc"        "map_dfr"        "map_if"         "map_int"       
[33] "map_lgl"        "map_raw"        "map2"           "map2_chr"      
[37] "map2_dbl"       "map2_df"        "map2_dfc"       "map2_dfr"      
[41] "map2_int"       "map2_lgl"       "map2_raw"       "pmap"          
[45] "pmap_chr"       "pmap_dbl"       "pmap_df"        "pmap_dfc"      
[49] "pmap_dfr"       "pmap_int"       "pmap_lgl"       "pmap_raw"      

By contrast, in the base-R version, we indeed stuck to base-R! There are only four main functions to learn in the 'apply' family: apply(), lapply(), sapply() and tapply().

Tibbles

Similarly, it is bad pedagogy to force students to learn tibbles, a more complex technology, instead of data frames, a simpler one. The types of situations that tibbles are meant to address should be an advanced topic, not for beginners with no coding background.

Those advanced situations involve data frames in which row/column elements are not atomic objects, i.e.\ not simple numbers, character strings or logical values. This is a "straw man" set up by the Tidy advocates'; R beginners are very unlikely to encounter data frames of this type.

The English issue

Again, the point most emphasized by Tidyverse advocates is that the Tidyverse is more teachable because of its "English-like" syntax.

Below is a comparison of the "English" dplyr to the "non-English" data.table (adapted from here): We'll again use R's built-in mtcars dataset.

mtdt 

Is there really any difference? Can't beginners, even without programming background, quickly adapt to either one after seeing a few examples? Even those who claim high teachability for dplyr do readily agree that their students could also easily pick up data.table, or for that matter my preference for beginners, base-R, given some examples.

We saw earlier that the purrr example,

mtcars %>%
  split(.$cyl) %>%
  map(~ lm(mpg ~ wt, data = .)) %>%
  map(summary) %>%
  map_dbl("r.squared")

would be baffling even to experienced (but non-R) programmers, starkly contradicting the claimed "English-like clarity" of Tidy. And the dplyr meaning of mutate is nowhere near its English meaning, and again, would not be guessed even by a non-R professional programmer.

In other words, though students may say they like the "English" aspect of Tidy, the benefit is illusory. They could become more proficient in R, more quickly, learning base-R than Tidy.

By the way, as noted below, the Tidy advocates don't like the many base-R functions whose names do use English, e.g. plot(), lines(), aggregate() and merge(). Clearly, then, English is not the core issue.

Pipes

The Tidyverse also makes heavy use of magrittr pipes, e.g. writing the function composition h(g(f(x))) as

f(x) %>%  g() %>% h()

Again, the pitch made is that this is "English," in this case in the sense of reading left-to-right. But again, one might question just how valuable that is, and in any event, I personally tend to write such code left-to-right anyway, without using pipes:

a 

As a long-time teacher of programming languages (C, C++, Java, Pascal, Python, R, assembly language, etc.), I find the promotion of pipes troubling. The piped version hides the fact that g() and h() have an argument, which is invisible in the pipe expression.

Or if w() say, were to have two arguments, the first one being used in the pipe, that argument would be hidden, making it appear that there is only one argument:

> w  3 %>% w(5)
[1] 13

Here w() has 2 arguments, but it looks like 1.

And what if we want that 3 to play the role of v, not u? Yes, magrittr has a way to do that, the "dot" notation:

> 3 %>% w(5,.)
[1] 11

But that is yet another example of my point, that Tidy is burdening the R learner with extra, unnecessary complexity. Indeed, just as dplyr, with 263 functions, is far too complex for beginners, so are pipes. There are so many variations to learn that Hadley's R for Data Science book devotes a full chapter to pipes, 3415 words.

As noted before, a beginner need learn only a small fraction of that material at first, but the above example of the dot notation is certainly not an advanced case. Again, each time the beginner is confronted with a new situation, she must sift through the myriad variants, of dplyr, purrr, pipes or whatever.

Moreover, what if the function h() above has two arguments, rather than just one, with each argument requiring functional composition? Pipes can't be used there.

And even more importantly, even advocates of pipes concede that pipes make debugging more difficult; by contrast, my style above lends itself easily to debugging. And again, for large problems, piped code is slower.

The benefit of pipes claimed by the Tidyers is the "left to right" execution. They conceded that one can achieve this without pipes, but stress that this comes at the expense of setting up variables for the intermediate results. That's a valid point, but is that small gain worth the increased cognitive load on R learners, increased debugging problems and so on? To me it clearly is not.

Code readability

The Tidyverse advocates also claim that the "English" in dplyr makes the code easier to read, not just write. To me, that is missing the point; as any instructor of software engineering can tell you, the best way to make code readable is to use REAL English, in good, meaningful code comments. And this is just as important, if not more so, for nonprogrammers.

See my R style guide for more on readability issues.

Summary: the proper status of the Tidyverse in teaching

As I said earlier, in discussions with those who report success in using the Tidyverse to teach beginning programmers, I ask whether their students are incapable of learning just base-R. They readily concede that the answer is no. Indeed, before the Tidyverse, throngs of people were learning base-R without any prior programming background.

As also mentioned, the Tidyverse can be difficult to debug, and run very slowly on large datasets. The perceived benefit of being "English-like" is illusory.

In short, in my view there is no advantage to teaching R through the Tidyverse, and some significant disadvantages. I think it is a mistake to feature the Tidyverse in teaching R to beginners, for these reasons:

  1. Complexity and volume of the presented material.

  2. Difficulty in debugging.

  3. Inadequate generalizability.

RECOMMENDATIONS

Teaching:

Courses in R, especially those aimed an nonprogrammers, should develop a solid grounding in base-R as first priority.

The proper placement of Tidy in R courses should be:

  • dplyr: Taught, along with data.table, at the Intermediate R level.

  • purrr: Taught only at the Advanced level.

  • pipes: Taught at the Intermediate level, and presented as an option that some students may find useful in some situations (as opposed to being presented as the way one should work).

I am certainly not saying one should only use base R; on the contrary, CRAN is a major advantage of R, which I use extensively, and to which R beginners should definitely be exposed.

But the Tidyverse should be considered advanced R, not for beginners, just as is the case for most complex CRAN packages, and should be presented, as noted, as an option, not as they way.

The role of RStudio:

In my view, RStudio can easily remedy the problem. It can take the following actions to greatly ameliorate the "monopolistic" problems:

  1. Promote the teaching of base-R to beginners, treating the Tidyverse as an advanced topic. The popular book, R for Everyone: Advanced Analytics and Graphics (second ed.), by Jared Lander does exactly this!

  2. In the various RStudio Web pages on writing fast R code, give data.table equal time.

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.