## The generic-random library, part 1: simple generic Arbitrary instances

(Note, 9/21/2017: The `generic-random` package has been changed and updated quite a bit since I wrote this post. For more up-to-date information, see e.g. the tutorial included with the package.)

In a previous post I pointed out that we know all the theory to make nice, principled, practical random generators for recursive algebraic data types; someone just needed to step up and do the work. Well, Li-yao Xia took up the challenge and produced a brilliant package, generic-random, available on Hackage right now for you to use!

However, although the package does include some Haddock documentation, it is probably difficult for someone with no experience or background in this area to navigate. So I thought it would be worth writing a few blog posts by way of a tutorial and introduction to the package.

``````> {-# LANGUAGE GADTSyntax           #-}
> {-# LANGUAGE DeriveGeneric        #-}
> {-# LANGUAGE FlexibleContexts     #-}
> {-# LANGUAGE UndecidableInstances #-}
>
> import GHC.Generics
> import Test.QuickCheck
>
> import Generic.Random.Generic
``````

## The problem

First, a quick recap of the problem we are trying to solve: the obvious, naive way of generating random instances of some recursive algebraic data type often produces really terrible distributions. For example, one might generate really tiny structures most of the time and then occasionally generate a humongous one. For more background on the problem, see this post or this one.

## A first example: generating generic `Arbitrary` instances

As a first example, consider the following algebraic data type:

``````> data Foo where
>   Bar  :: Char -> Int -> String -> Foo
>   Baz  :: Bool -> Bool -> Foo
>   Quux :: [Woz] -> Foo
>   deriving (Show, Generic)
>
> data Woz where
>   Wiz :: Int -> Woz
>   Waz :: Bool -> Woz
>   deriving (Show, Generic)
``````

You have probably noticed by now that this is not recursive (well, except for the embedded lists). Patience! We’ll get to recursive ADTs in due time, but it turns out the library has some nice things to offer for non-recursive ADTs as well, and it makes for an easier introduction.

Now, suppose we wanted to use QuickCheck to test some properties of a function that takes a `Foo` as an argument. We can easily make our own instances of `Arbitrary` for `Foo` and `Woz`, like so:

``````instance Arbitrary Foo where
arbitrary = oneof
[ Bar <\$> arbitrary <*> arbitrary <*> arbitrary
, Baz <\$> arbitrary <*> arbitrary
, Quux <\$> arbitrary
]

instance Arbitrary Woz where
arbitrary = oneof
[ Wiz <\$> arbitrary
, Waz <\$> arbitrary
]``````

This works reasonably well:

``````λ> sample (arbitrary :: Gen Foo)
Baz True True
Baz False True
Baz True True
Quux []
Baz False True
Bar '<' 3 "zy\\\SOHpO_"
Baz False True
Bar '\SOH' 0 "\"g\NAKm"
Bar 'h' (-9) "(t"
Quux [Wiz (-2),Waz False]
Baz False True``````

The only problem is that writing those instances is quite tedious. There is no thought required at all. Isn’t this exactly the sort of thing that is supposed to be automated with generic programming?

Why yes, yes it is. And the `generic-random` package can do exactly that. Notice that we have derived `Generic` for `Foo` and `Woz`. We can now use the `genericArbitrary` function from `Generic.Random.Generic` to derive completely standard `Arbitrary` instances, just like the ones we wrote above:

``````> instance Arbitrary Foo where
>   arbitrary = genericArbitrary
>
> instance Arbitrary Woz where
>   arbitrary = genericArbitrary
``````
``````λ> sample (arbitrary :: Gen Foo)
Quux []
Bar '\159' (-2) ""
Baz True True
Baz False False
Baz True True
Baz True False
Quux [Wiz 9,Wiz 7,Waz True,Waz True,Waz False]
Quux [Wiz (-10),Waz False,Waz False,Waz True,Waz True,Wiz (-14),Wiz 13,Waz True,Wiz (-8),Wiz 12,Wiz (-13)]
Bar '\130' 10 "FN\222j?\b=\237(\NULW\231+ts\245"
Bar 'n' 14 ""
Bar '\205' 4 "\SYN"``````

Seems about the same, except we wrote way less code! Huzzah!

If we want certain constructors to occur more frequently, we can also control that using `genericArbitraryFrequency`, which takes a list of `Int`s (each `Int` specifies the weight for one constructor).

A few notes:

• Using the `Generic.Random.Generic` module is the quickest and simplest way to generate random instances of your data type, if it works for your use case.

• It has some limitations, namely:

• It only generates `Arbitrary` instances for QuickCheck. It can’t create more general random generators.

• It probably won’t work very well for recursive data types.

However, these limitations are addressed by other parts of the library. Intrigued? Read on!

## Recursive types, the simple way

Let’s now consider a simple recursive type:

``````> data Tree a where
>   Leaf   :: a                -> Tree a
>   Branch :: Tree a -> Tree a -> Tree a
>   deriving (Show, Generic)
>
> treeSize :: Tree a -> Int
> treeSize (Leaf _)     = 1
> treeSize (Branch l r) = 1 + treeSize l + treeSize r
``````

We can try using `genericArbitrary`:

``````instance Arbitrary a => Arbitrary (Tree a) where
arbitrary = genericArbitrary``````

The problem is that this tends to generate some tiny trees and some enormous trees, with not much in between:

``````λ> map treeSize  replicateM 50 (generate (arbitrary :: Gen (Tree Int)))
[1,1,1,269,1,1,1,1,1,11,7,3,5,1,1,1,7,1,1,1,3,3,83,5,1,1,3,111,265,47,1,3,19,1,11,1,5,3,15,15,1,91,1,13,4097,119,1,15,5,3]``````

And this is not a problem specific to trees; this kind of thing is likely to happen for any recursive type.

Before we get to more interesting/complicated tools, it’s worth noting that `random-generics` provides a simple mechanism to limit the size of the generated structures: the `genericArbitrary'` function works like `genericArbitrary` but uses QuickCheck’s `sized` mechanism to cut off the recursion when it gets too big. The available size is partitioned among recursive calls, so it does not suffer from the exponential growth you might see if only the depth was limited. When the size counter reaches zero, the generator tries to terminate the recursion by picking some finite, non-recursive value(s). The parameter to `genericArbitrary'` is a natural number specifying how deep the finite, recursion-terminating values can be. `Z` (i.e zero) means the generator will only be willing to terminate the recursion with nullary constructors. In our case, `Tree` does not have any nullary constructors, so we should not use `Z`: if we do, the generator will be unable to terminate the recursion when the size reaches zero and we will get the same behavior as `genericArbitrary`. Instead, we should use `S Z`, which means it will be able to pick the depth-1 term `Leaf x` (for some arbitrary `x`) to terminate the recursion.

Let’s try it:

``````> instance (Arbitrary a, Generic a, BaseCases Z (Rep a)) => Arbitrary (Tree a) where
>   arbitrary = genericArbitrary' (S Z)
``````
``````λ> sample (arbitrary :: Gen (Tree Int))
Leaf 0
Branch (Leaf 0) (Branch (Leaf 0) (Branch (Leaf 0) (Leaf 0)))
Branch (Leaf (-1)) (Leaf 1)
Leaf (-3)
Leaf 7
Branch (Leaf (-4)) (Branch (Branch (Leaf 1) (Leaf (-1))) (Leaf (-1)))
Branch (Leaf (-2)) (Branch (Leaf 1) (Branch (Leaf 0) (Branch (Leaf 0) (Leaf 0))))
Leaf 14
Branch (Branch (Leaf 2) (Leaf 2)) (Branch (Branch (Branch (Leaf 1) (Branch (Branch (Leaf 0) (Branch (Leaf 0) (Leaf 0))) (Branch (Leaf 0) (Leaf 0)))) (Branch (Branch (Branch (Leaf 0) (Leaf 0)) (Leaf 0)) (Leaf 0))) (Leaf (-3)))
Leaf 4
Leaf 9``````

Ah, that’s much better.

Finally, `genericArbitraryFrequency'` is the same as `genericArbitraryFrequency` but limits the recursion depth as `genericArbitrary'` does.

If you have a recursive data type you want to use with QuickCheck, it’s worth trying this, since it is quick and simple. The main problem with this approach is that it does not generate a uniform distribution of values. (Also, it is limited in that it is specifically tied to QuickCheck.) In this example, although you can’t necessarily tell just by looking at the sample random trees, I guarantee you that some kinds of trees are much more likely to be generated than others. (Though I couldn’t necessarily tell you which kinds.) This can be bad if the specific trees that will trigger a bug are in fact unlikely to be generated.

Next time, we’ll look at how we can actually have efficient, size-limited, uniform random generators using Boltzmann samplers.

Advertisements
Posted in combinatorics, haskell | Tagged , , , | 4 Comments

## Meeting people at ICFP in Nara

In less than 24 hours I’m getting on a plane to Japan (well, technically, Dallas, but I’ll get to Japan eventually). As I did last year, I’m making an open offer here: leave a comment on this post, and I will make a point of finding and meeting you sometime during the week! One person took me up on the offer last year and we had a nice chat over dinner.

Posted in meta | Tagged | 6 Comments

## Deep work and email habits

Lately I have been enjoying Cal Newport’s writing on work, and particularly his new book Deep Work which I am in the middle of reading (definitely recommended). His basic thesis is about the power of sustained, focused, distraction-free work on cognitively demanding tasks—what he calls deep work. It takes intentional effort to make the time and space for this kind of work, but Newport argues cogently that doing so can have enormous benefits.

Newport’s ideas have really resonated with me—I think I was already converging (albeit slowly, with little clarity) on similar ideas and practices over the last few years—and I’ve begun trying to put some of them more deliberately into practice. First, I have scheduled two large (4 hour) blocks of time for deep work each week. These blocks are sacrosanct: I won’t meet with students, schedule committee meetings, or do anything else during those times. I physically go somewhere other than my office—usually the library, occasionally my favorite coffee shop, somewhere relatively quiet and interruption-free where students and colleagues won’t find me. I first do as much as possible without turning on my laptop: course planning, reading, brainstorming, a lot of longhand writing (blog posts, papers, grant proposals, whatever—for example, I wrote this blog post itself longhand during my deep work session this morning). Sometimes if I need to write a longer, thoughtful email response, I will even print out the message beforehand and write my response longhand. Only towards the end of the session will I pull out my laptop, if I have specific projects to work on deeply that require a computer, like some sort of coding project.

Anecdotally at least, so far this feels incredibly successful—I get a lot done during these deep work sessions and always come away feeling accomplished and energized. The thing that feels especially good is that I’m not just getting a large amount of stuff done, but I’m getting important, difficult stuff done.

Another related practice I have recently adopted is that I do not read or write any email before 4pm. I have literally blocked myself from accessing email on my computers and phone between midnight and 4pm. Perhaps this sounds heretical, but that’s just the point—“because doing otherwise would be heresy” is a terrible reason for doing anything, and the fact is that not many of us really stop to consider and consciously choose the way we make use of technologies like email and social media. It’s taken some getting used to, but by now I don’t think I am ever going back. At 4pm I triage my inbox—respond to things that need a quick response, archive or delete stuff I don’t care about, and forward other things to my personal bug tracker for dealing with later. I am typically able to totally clear out my inbox before going home for the day. Over the course of the day I keep a list of emails I want to write later, and I write those at the same time that I triage my inbox, or sometimes later in the evening before going to bed. It feels way more efficient to batch most of my email processing into a focused session like this, and freeing to not be distracted by it the rest of the day. But do I ever miss it? Yes, all the time—and that’s exactly the point! Left to my natural tendencies I distract myself silly checking my email constantly.

Time will tell how much of this sticks and how my approach might change over time—I’ve scheduled a reminder for myself to write a followup post six months from now. As always, I’m happy to hear and respond to thoughts, reactions, questions, etc. in the comments.

Posted in meta | Tagged , , | 8 Comments

## Academic integrity: context and concrete steps

Continuing from my previous post, I wanted to write a bit about why I have been thinking about academic integrity, and what, concretely, I plan to do about it.

So, why have I been thinking about this? For one thing, my department had its fair share of academic integrity violations last year. On the one hand, it is right for students to be held accountable for their actions. On the other, in the face of a spate of violations, it is also right for us to reevaluate what we are doing and why, what sort of environmental factors may be pushing students to violate academic integrity, and how we can create a better environment. Environment does not excuse behavior, but it can shape behavior in profound ways.

Another reason for thinking about academic integrity is that starting this fall, I will be a member of the committee that hears and makes a determination in formal academic integrity cases at my institution. It seems no one wants to be on this committee, and to a certain extent I can understand why. But I chose it, for several reasons. For one, I think it is important to have someone on the committee from the natural sciences (I will be the only one), who understands issues of plagiarism in the context of technical subjects. I also care a lot about ensuring that academic integrity violations are handled carefully and thoughtfully, so that students actually learn something from the experience, and more importantly, so that they come through with their sense of belonging intact. When a student (or anyone, really) does something that violates the standards of a community and is subject to consequences, it is all too easy for them to feel as though they are now a lesser member or even excluded from the community. It takes much more intentional communication to make clear to them that although they may have violated a community standard—which necessarily comes with a consequence—they are still a valued member. (Thanks to Leslie Zorwick for explaining about the power of belonging, and for relating recent research showing that communicating belonging can make a big difference for students on academic probation—which seems similar to students accused or convicted of academic integrity violations. I would cite it but I think it is not actually published yet.)

Thinking about all of this is well and good, but what will I do about it? How do I go about communicating all of this to my students, and creating the sort of environment I want? Here are the concrete things I plan to do starting this fall:

• In all my courses where it makes sense, I plan to require students to have at least one citation (perhaps three, if I am bold) on every assignment turned in—whether they cite web pages, help from TAs or classmates, and so on. The point is to get them thinking regularly about the resources and help that they make use of on every single assignment, to foster a spirit of thankfulness. I hope it will also make it psychologically harder for students to plagiarize and lie about it. Finally, I hope it will lead to better outcomes in cases where a student makes inappropriate use of an online resource—i.e. when they “consult” a resource, perhaps even deceiving themselves into thinking that they are really doing the work, but end up essentially copying the resource. If they don’t cite the resource in such a case, I have a messy academic integrity violation case on my hands; if they do, there is no violation, even though the student didn’t engage with the assignment as I would have hoped, and I can have a simple conversation with them about my expectations and their learning (and perhaps lower their grade).

• I will make sure to communicate to my students how easy it is for me to detect plagiarism, and how dire the consequences can be. A bit of healthy fear never hurt!

• But beyond that, I want to make sure my students also understand that I care much more about them, as human beings, than I do about their grade or whether they turn in an assignment. I suspect that a lot of academic integrity violations happen at 2am, the night before a deadline, when the student hasn’t even started the assignment and they are riddled with anxiety and running on little sleep—but they feel as though they have to turn something in and this urge overrides whatever convictions they might have about plagiarism. To the extent their decision is based on anxiety about grades, there’s not much I can do about it. However, if their decision stems from a feeling of shame at not turning something in and disappointing their professor, I can make a difference: in that moment, I want my students to remember that their value in my eyes as human beings is not tied to their academic performance; that I will be much more impressed by their honesty than by whether they turn something in.

• As a new member of the academic integrity committee, I plan to spend most of my time listening and learning from the continuing members of the committee; but I do hope to make sure our communication with both accused and convicted students emphasizes that they are still valued members of our community.

Other concrete suggestions, questions, experiences to relate, etc. are all most welcome!

Posted in teaching | | 2 Comments

## Academic integrity and other virtues

I have been thinking a lot recently about academic integrity. What does it mean? Why do we care—what is it we fundamentally want students to do and to be? And whatever it is, how do we go about helping them become like that?

As a general principle, I think we ought to focus not just on prohibiting certain negative behaviors, but rather on encouraging positive behaviors (which are in a suitable sense “dual” to the negative behaviors we want to prohibit). Mere prohibitions leave a behavioral vacuum—“OK, don’t do this, so what should I do?”—and incentivize looking for loopholes, seeing how close one can toe the line without breaking the letter of the law. On the other hand, a positive principle actively guides behavior, and in actively striving towards the ideal of the positive principle, one (ideally) ends up far away from the prohibited negative behavior.

In the case of academic integrity, then, it is not enough to say “don’t plagiarize”. In fact, if one focuses on the prohibition itself, this is a particularly difficult one to live by, because academic life is not lived in a vacuum: ideas and accomplishments never spring forth ex nihilo, owing nothing to the ideas and accomplishments of others. In reality, one is constantly copying in big and small ways, explicitly and implicitly, consciously and unconsciously. In fact, this is how learning works! We just happen to think that some forms of copying are acceptable and some are not. Now, there are good reasons for distinguishing acceptable and unacceptable copying; the point is that this is often more difficult and ambiguous for students than we care to admit.

So what is the “dual” of plagiarism? What are the positive virtues which we should instill in our students? One can, of course, say “integrity”, but I don’t think this goes far enough: to have integrity is to adhere to a particular set of moral principles, but which ones? Integrity means being truthful, but truthful about what? It seems this is just another way of saying “don’t plagiarize”, i.e. don’t lie about the source of an idea. I have come up with two other virtues, however, which I think really get at the heart of the issue: thankfulness and generosity. (And in the spirit of academic thankfulness, I should say that Vic Norman first got me thinking along these lines with his paper How Will You Practice Virtue Witout Skill?: Preparing Students to be Virtuous Computer Programmers, published in the 2014-2015 Journal of the ACMS; I was also influenced by a discussion of Vic’s paper with several others at the ACMS luncheon at SIGCSE 2016.)

Academic thankfulness has to do with recognizing one’s profound debt to the academic context: to all those thinkers and doers who have come before, and to all those who help you along your journey as a learner, whether professors, other students, or random strangers on the Internet. A thankful student is naturally driven to cite anything and everything, to give credit where credit is due, even to give credit where credit is not technically necessary but can serve as a token of thanks. A thankful student recognizes the hard work and unique contributions of others, rather than seeing others as mere means to their own ends. A thankful student never plagiarizes, since taking something from someone else and claiming it for one’s own is the height of ingratitude.

Academic generosity is about freely sharing one’s own ideas, sacrificing one’s time and energy to help others, and allowing others to share in credit and recognition. Being academically generous is harder than being thankful, because it opens you up to the potential ingratitude of others, but in some sense it is the more important of the two virtues: if no one were generous, no one would have anything to be thankful for. A generous student is naturally driven to cite anything and everything, to give credit and recognition to others, whether earned or not. A generous student recognizes others as worthy collaborators rather than as means to an end. A generous student never plagiarizes, since they know how it would feel to have their own generosity taken advantage of.

There’s more to say—about the circumstances that have led me to think about this, and about how one might actually go about instilling these virtues in students, but I think I will leave that for another post.

Posted in teaching | | 1 Comment

## POGIL workshop

A few weeks ago I attended a 3-day training workshop in St. Louis, put on by the POGIL project. I attended a short POGIL session at the SIGCSE CS education conference in March and was sufficiently impressed to sign up for a training workshop (it didn’t hurt that Clif Kussmaul has an NSF grant that paid for my registration and travel).

POGIL is an acronym for “Process Oriented Guided Inquiry Learning”. Process-oriented refers to the fact that in addition to learning content, an explicit goal is for students to learn process skills like analytic thinking, communication, and teamwork. Guided inquiry refers to the fact that students are responsible for constructing their own knowledge, guided by carefully designed questions. The entire framework is really well thought-out and is informed by concrete research in pedagogical methods. I really enjoyed how the workshop used the POGIL method to teach us about POGIL (though of course it would be rather suspect to do anything else!). It gave me not just an intellectual appreciation for the benefits of the approach, but also a concrete understanding of the POGIL experience for a student.

The basic idea is to put students in groups of 3 or 4 and have them work through an activity or set of questions together. So far this sounds just like standard “group work”, but it’s much more carefully thought out than that:

• Each student is assigned a role with specific responsibilities within their group. Roles typically rotate from day to day so each student gets a chance to play each role. Roles can vary but common ones include things like “manager”, “recorder”, “reporter”, and so on. I didn’t appreciate how important the roles are until attending the workshop, but they are really crucial. They help ensure every student is engaged, forestall some of the otherwise inevitable social awkwardness as students figure out how to relate to their group members, and also play an important part in helping students develop process skills.

• The activities are carefully constructed to take students through one or more learning cycles: beginning with some data, diagrams, text, etc. (a “model”), students are guided through a process starting with simple observations, then synthesis and discovering underlying concepts, and finally more open ended/application questions.

The teacher is a facilitator: giving aid and suggestions as needed, managing dificulties that arise, giving space and time for groups to report on their progress and share with other groups, and so on. Of course, a lot of work goes into constructing the activities themselves.

In some areas, there is already a wealth of POGIL activities to choose from; unfortunately, existing materials are a bit thinner in CS (though there is a growing collection). I won’t be able to use POGIL much this coming semester, but I hope to use it quite a bit when I teach algorithms again in the spring.

Posted in teaching | Tagged , , , , | 1 Comment

## New Haskell Symposium paper on “twisted functors”

Satvik Chauhan, Piyush Kurur and I have a new paper which will appear at the 2016 Haskell Symposium in Japan:

How to Twist Pointers without Breaking Them

Although pointer manipulations are used as a primary motivating example, at heart the paper is really about “twisted functors”, a class of applicative functors which arise as a natural generalization of the semi-direct product of two monoids where one acts on the other. It’s a really cute idea1, one of those ideas which seems “obvious” in retrospect, but really hadn’t been explored before.

We give some examples of applications in the paper but I’m quite certain there are many other examples of applications out there. If you find any, let us know!

1. I can say that since it wasn’t actually my idea!

Posted in haskell, writing | | 6 Comments