Rick Winfrey

30 Days of Haskell

May 25, 2015 » 4 minutes (731 words)

I recently attended Lambda Conf 2015. Coming from a dynamically typed background of working mostly in Ruby and Clojure, I had only a superficial understanding of type theory and how it applies to functional programming in languages like Haskell and Purescript. I had some vague idea that static typing in a functional programming context would result in better, easier to reason about, and easier to maintain code, but I hadn't experienced it enough first-hand to confidently say yes to these claims. I'd like to clarify that I still have a superficial understanding of type theory, however, the certainty and confidence one gains by working in a statically typed language is very different than dynamically typed languages, even if those dynamically typed languages are functional like Clojure. The confidence and certainty of type safety at compile time provides guarantees for free that I would otherwise have to write many tests to achieve. If for no other reason, I see statically typed functional programming as the way forward because it results in less code, is more resilient to change, and provides a near universal language for describing very abstract ideas with precision and brevity.

The big turning point in my thinking was a result of hearing Chris Allen describe an experience of debugging some problematic code in Clojure. A 250 line namespace was buggy, and it wasn't clear why. All the tests were passing, but debugging took a few developers and a few hours, and numerous `println` statements to understand that malformed data was propagating through the code due to vector implementing Clojure's IFn interface. I have personally experienced the pain of debugging Clojure projects whose tests pass, but errors abound at runtime. And I can attest that in all of the runtime errors I have debugged in Clojure, the heart of the problem was a type issue. All of those issues could have been easily avoided had the functions had a compile time type check. I can also attest that a majority of the tests I write for Clojure code deal with directly or indirectly verifying input and output types, with few tests validating the correctness of business logic.

Throughout the conference, I learned of many new ideas and terms, usually related to type theory. After the conference, I feel inspired to devote a chunk of each day to learning more about Haskell and Purescript, and want to find a balance between working in and exploring those languages, while also reading and studying Haskell Programming From First Principles and the various papers it suggests for further reading. Additionally, I'd like to try my hand at studying Category Theory and refresh my understanding of Proof Theory, although that is a stretch goal. The reason for refreshing Proof Theory is to prepare for potentially exploring Idris, a dependently typed functional programming language that allows for some degree of specifying logic in the types themselves, which then allows the programmer to "prove" the correctness of a program by making "propositions" about expressions involving those types, and then showing that the propositions hold true using a variety of proof tactics such as induction. I have but the most simple and superficial understanding of Idris, but in seeing a demonstration of its power via a workshop at Lambda Conf, Idris sticks out to me as a potential future for statically typed functional programming.

My goal is simple: learn as much as I can about Haskell and its many related concepts and ideas before I begin a graduate program in Predictive Analtics at Depaul beginning in 40 days. My intuition tells me that Haskell's type system would lend itself well to various types of data analysis and machine learning techniques, however the core instruction will likely be in languages like R and Python. Considering that most data analysis is a series of transformations of the raw data, so that it can be easily processed and reduced or mapped over with various functions, the domain of data analysis lends itself very well to functional programming. Ideally I would love to come out of formally studying data analysis and statistics with a strong ability in applying these techniques in Haskell or Purescript, and if I'm able to make good headway, perhaps can also prove the correctness of data analysis programs using Idris...

Happy λ!