Understand why Transducers can Improve Performance

Paul Frend
InstructorPaul Frend
Share this video with your friends

Social Share Links

Send Tweet
Published 7 years ago
Updated 5 years ago

Chaining operations like map and filter on an array can have performance issues as we have to iterate through the whole array for every operation.

In this lesson we explore this problem by doing some basic array transformations and measuring their performance.

We’ve got two utility function to help us with this:

  • arrayOfRandoms is a higher order function which helps us create a range of random numbers
  • timeIt will help us measure how long functions take to execute

It's important to understand this problem and why it occurs before we move on to how transducers handle it differently.

To understand why transducers are useful, we first have to understand the problem of chaining operations when iterating through arrays. Let's do some performance tests. To help us with that, we've got this arrayOfRandoms function, which will just help us generate arrays of various lengths with random numbers up to this random ceiling.

We've also got this timer function which will just help us time how long functions take to execute. Let's use arrayOfRandoms to create two arrays. The first one will have 1,000 values in it, and the second one will have one million values.

Just a quick tip, if you haven't seen this 1E6 syntax before, it's a really handy way to define large numbers, since you won't go crazy counting zeros. It's called a scientific notation literal, and it means to multiply a number by 10, then raising to a given power.

1E6 would be the same as one times ten to the power of six, but you can think of it simply as how many zeros do I want to add? Now, let's get to work with some array operations. I'm going to write const resultFrom1000, and let's start working on our array of thousands.

Let's say a business requirement is to triple our values, and then only include those that are even. Let's add on a map, and that's going to return a value times three. A filter operation is a perfect match for our second requirement, so let's just put this on a new line.

Then we're going to call filter. Our predicate is only going to return true for even values. We'll be using these functions a few times, so let's extract them. Let's call the first one isEven, and that will be the result of our filter predicate. We'll call the second one tripleIt. That will our map call.

Let's see how this performs. Let's start by only measuring the call to map. We'll label this thousand-map, and it'll take an anonymous function. In here, we just copy in our execution, but we'll delete the filter.

Let's copy this again, and we'll name this mapAndFilter. We'll just add back our call to filter in here. Let's see how they compare. We see that our call to map took 0.6 milliseconds, but our call to mapAndFilter took 1.8.

Now, depending on your app, this is usually nothing you have to worry about, as these numbers are really small anyway. No real alarm bells yet, but let's do the same thing to one million records. Let's just copy both of these blocks, and we'll just rename thousand to million. We'll change these to use our array of millions inside.

Now, let's compare these two. As we can see, our call to mapAndFilter took about 1.3 seconds, but our call to map took only 750 milliseconds. That's getting up there where users will definitely start noticing the difference.

What can we do? We still have to have our two operations. The problem is that both map and filter have to go through the whole array. We're iterating through the collection twice. What I mean by that is a call to map will produce a new array with all the values here as an intermediary collection. Then filter will go through that array again from the beginning.

How can we keep all our logic, but only iterate through the collection once? Generally, we solve this by falling bad on our good old imperative problem solving skills. Let's give that a go. Let's just do in-line within one of these timing functions.

Let's call this timeItMillionImperative, and we don't need to save our results. We're just interested in the outcome. Instead of mapping and filtering, let's create an external variable to hold our result, which will be an empty array.

Then we'll comment our map and filter, and instead, we're going to call a forEach, which will take a value. Then in here, we want to replicate our call to map. Let's create a const called tripled, which will be the result of calling tripleIt with our value.

Then for our filtering operation, we just want to check if our tripled value is even. If it is, we want to push it onto our result. Now, let's measure this. Our execution time is now down to 344 milliseconds, compared to 1,400, but our code quality has suffered.

Wouldn't it be great if we could keep our array operations defined separately, but still only iterate once through our collection? This is where transducers are going to save us from our misery.

press0
press0
~ 6 years ago

link to code

Paul Frend
Paul Frendinstructor
~ 6 years ago

@press0 we're in the process of adding the sample code.

Adam
Adam
~ 6 years ago

@Paul, what is the tool you are using for inline evaluation?

Antonio Sejas
Antonio Sejas
~ 6 years ago

@paul what code editor do you use ?? Thank you ! ;)

Adam
Adam
~ 6 years ago

Sublime text 3

Adam Bertrand

Paul Frend
Paul Frendinstructor
~ 6 years ago

@Adam, sorry for the late reply. The inline evaluation is using Quokka

@Antonio I'm using Webstorm here with all the extra toolbars hidden

Markdown supported.
Become a member to join the discussionEnroll Today