[英文版]Adit Bhargava：Illustrations break the spell on Algorithm（圖靈訪談）

訪談嘉賓：

（Adit自畫像）

Adit Bhargava， 軟體工程師，兼具電腦科學和美術方面的教育背景，在adit.io撰寫程式設計方面的部落格。

因為愛好，Adit踏入了程式設計殿堂。Visual Basic 6 for Dummies教會了他很多基礎知識，但始終對演算法沒搞明白。直到遇到一位優秀的演算法教授後，他才認識到這些概念是多麼地簡單且優雅。

幾年前，他開始在adit.io上撰寫圖解式博文，介紹函數語言程式設計、Git、機器學習和併發。圖解式寫作風趣幽默、化繁為簡、清晰明確，受到大量讀者的喜愛。

---

我們主要聊了些：

• 為什麼要寫這樣一本萌萌的演算法入門書
• 封面插畫背後的故事
• Adit神祕的演算法導師
• Adit最喜歡的演算法
• 評判演算法的重要指標
• 程式設計學習低齡化

Chinese Version

Why would you like to write such an introductory book, which is full of fascinating scenarios and cute illustrations drawn by hand?

I usually take notes for myself when I learn something, because it helps me learn. For example here are the notes I'm taking as I read "A Book Of Abstract Algebra" (attached). So this was a technique I had used in the past, and I thought others might find it useful also. So I wrote this blog post. People liked it and it made me think that a book with the same style would probably do pretty well too.

（Let's enjoy Adit's explanation of Monad in pictures！）

學習 Monad的渠道：

• 取得電腦科學專業的博士學位。
• 壓根兒不學。這裡根本用不到那些條條框框！

Functor 將一個普通函式應用到被封裝的值上：

Applicative 將一個封裝的函式應用到封裝的值上：

Monad 將一個 “接受普通值並回傳一個被封裝值” 的函式應用到一個被封裝的值上，這一任務由函式 >>=（讀作 bind）完成。聽起來似乎很拗口，讓我們來看個例子吧，還是熟悉的 Maybe：

假設 half 是隻對偶數感興趣的函式：

half x = if even x
     then Just (x `div` 2)
     else Nothing

如果扔給 half 一個封裝的值會怎樣？

這時，我們需要用 >>= 把被封裝的值擠到 half中。猜猜>>= 的照片：

再看看它的效果：

> Just 3 >>= half
Nothing
> Just 4 >>= half
Just 2
> Nothing >>= half
Nothing

這其中究竟發生了什麼？Monad 是另一種型別類，這是它定義的一部分：

class Monad m where
    (>>=) :: m a -> (a -> m b) -> m b

下圖展示了 >>= 各個引數的意義：

下面的定義讓 Maybe 成了 Monad：

instance Monad Maybe where
    Nothing >>= func = Nothing
    Just val >>= func  = func val

來看看執行 Just 3 時發生了什麼：

如果傳入 Nothing 就更容易了：

你還可以把這些呼叫過程連起來，比如執行 Just 20 >>= half >>= half >>= half 會得到 Nothing：

太棒啦！

（Taken from Adit's articleFunctors, Applicatives, And Monads In Pictures）

From the book cover, I thought it might be full of hand-drawn illustrations about mice. It seems not, for there’re other images like sheep, birds, rabbits, diagrams. Why would you put that picture in front of the book?

I wish I had a good answer for you! The people at Manning chose the picture on the cover. Manning was generally good about giving me a lot of control over the book, but for the cover, they really fell in love with this image and chose to use it.

A whole bunch of readers are really curious about your algorithmic teacher mentioned in author's introduction part who made tough concepts become simple but elegant. Could you share some of his/her teaching methods?

Sure! Her most effective teaching method was stepping through an algorithm line by line. When something is hard, it is easier skip over it. But in order to learn, it is important to slow down at the hard parts. So for each algorithm she taught, she would slow down and go through the code line by line and explain what each line did. I tried to do the same thing in my book. In the section on recursion, for example, I walk through each line and show how the stack changes. I think having that level of detail is really important.

What's your favorite algorithm? Why does it give you such a deep impression?

I've mentioned how much I love graph algorithms a few times in the book. Graphs are a really amazing structure that show up absolutely everywhere. I feel like I'm able to solve so many problems just using graph algorithms. I recently went to lunch with a friend and someone at the lunch said "I bet I can teach you Category Theory in 15 minutes". I didn't know anything about Category Theory, but I knew graphs and abstract algebra, so it actually took him less than five minutes to explain Category Theory to me. At work, I've been able to automate some tedious tasks, all because I know how topological sort and breadth-first search work.

Sometimes, short operational time does not necessarily mean good performances. Except time, are there any other dimensions to judge algorithm?

Yes! I think ease of use is a pretty important metric. For example, there are plenty of machine learning techniques more advanced than KNN, but if you are just starting out with a problem, you might want to start with KNN even if you are a machine learning expert. If an algorithm is easy to think about, there are fewer places for bugs to hide. Once you start getting into more complicated algorithms like neural networks, if you run into a bug, it will take more time to figure out the cause because there are so many more moving parts, so more places for the bug to hide. When picking an algorithm and considering performance time, it is important to think about the performance of the programmer also! Easy to understand code is more maintainable and more likely to be bug-free.

There are actually a few teenagers who are reading your book. What do you think of learning algorithm from very early ages, like primary school ages or even earlier?

I think that makes a lot of sense. Programming is a way to be creative, just like painting or music. I learned programming pretty early and made video games and animations. The earlier you learn, the sooner you can work on your own projects!

Will you keep this up with a "Grokking" series covering other CS/Dev topics? Because we all love it.

I hope so! I need to think hard about what else I can write about :)

---

——See More

---

更多精彩，加入圖靈訪談微信！