Diagrams!

New in sci-kit learn 0.23.0

by Shawn Syms

Image credits: Diagram by the Noun Project; Pipeline by the Noun Project; scikit learn logo, scikit learn

Some folks have Record Store Day to get psyched about. For others, it's news of a Spice Girls reunion that sets their hearts aflame. For my dad and brother, it was the day that Star Trek II: The Wrath of Khan finally reached Niagara Square Cinemas (of course, my mom and I went to see Flashdance instead.)

With all these categories of people in that paragraph, it was starting to sound like a classification problem: Now I'm getting excited!

OK, if you don't already know what I'm talking about, it is something that is near and dear to the hearts of machine-learning nerds on all seven continents (do NOT argue with me on that point!): the newest release of scikit learn is out! Can I get an OMG? Thank you.

If you are reading this, I would bet money (if I had some, I'm a poor student LOL!) that you already know what scikit learn is. In their own words, "Scikit-learn is an open source machine learning library that supports supervised and unsupervised learning. It also provides various tools for model fitting, data preprocessing, model selection and evaluation, and many other utilities. " Now, that is what I call a drab "about us" statement. Unlike me, these guys have clearly not worked in marketing before. But suffice it to say, scikit learn is one of the most popular machine-learning libraries out there in use by data scientists and others. Machine learning: you know what that is, right? Like, Alexa. You know. Stuff like that. Most important, machine learning is something that you do not need a PhD in order to either understand or implement.

Scikit-learn doesn't do everything. But it does a lot of things. And it's great to have such a handy toolkit of utilities that all play nicely together.

(You know, I thought nobody actually read these posts [except maybe for my instructor at General Assembly Toronto]. But I was wrong! I received feedback on this blog just the other day from a reader. To protect the person's privacy, I will only identify them as K.P. from Rochester, NY. K.P.'s sentiment was not only that my blog is good (which was great for my self-esteem because I'd thought it was sort of terrible until that moment), but particularly that my updates on the latest new features in sci-kit learn are both timely and information. So there you go. And here you go: another sci-kit learn update!)

sci-kit learn is totally to diagram for

OK, so scratch that about timely. There is another blog post out by someone else about the top 9 new features in sci-kit learn, even though the new release was days ago. I will spare you a link to it, even though it's a great article actually, because he is my competition. Google is your friend.

That keener wrote about it the very next day after the release. Well, unlike him, I have a life (right now, it is spent studying and attending classes about 12 hours a day, no joke). And I've been spending my days not on the cutting edge, working with boring old sci-kit learn features like the TF-IDF Vectorizer. (I'm tempted to show you the hand-drawn diagram I did of my current machine-learning project using TF-IDF vectorizer. But I won't. Because in it, I misspelled the term "cosine" as if it meant to "co sign" for a loan. Which is pretty humiliating (and I already have a chip on my shoulder about certain aspects of math!).

And, again unlike that other blogger, I'm not going to tell you about the top nine things in the new release (call me a cynic, but you've heard that everyone is supposed to write "listicles," because more people will click on digital articles that are written in the form of a list). I'm just going to write about the number one coolest thing in the new release. It would be the sci-kit learn version of the scene in Flashdance where, after she falls in the opening moments of her dance audition, she totally takes control of the room and wins everything. OK never mind, that scene is cooler than this; here, I'm just talking about some interactive diagrams. Big whoop.

But they are in fact really cool. And here are the Top Two Lines of Code that you. need. to. know. if you want to get yerself summa these here diagrams:

from sklearn import set_config

set_config(display='diagram')

Now, some of these model diagrams are pretty darn vanilla. An example or two:

Even this one for that aforementioned TF-IDF Vectorizer is a bit shoddy if you ask me (I think I like my hand-drawn, mistake-ridden diagram better):

That's a nice try at including my configuration parameters. I suppose it's my own fault they are truncated though, for using a custom stopwords set. The project is a recommender system for Canadian literature. Screaming headlines like "WINNER OF THE SCOTIABANK GILLER PRIZE" might look good in the Toronto Star or on posters at Chapters, but they confuse my model and make it want to recommend novels about Nova Scotia (about which there are quite a few good ones, actually).

But a place where this feature really shines is where you implement it to visualize complex pipelines of estimators and transformers. You know, when it would be tedious to sort through what happens when and the various parameter by scrolling through endless text outputs. I know you have been there.

I looked through some of my own recent projects to something to diagram for you. But in truth they were all a bit boring. Luckily the folks at scikit learn have kindly supplied a great example in their release notes. (You know I want to try and front like I'm on a first-name basis with those guys. Well, I know that one of them has the first name Olivier. I saw him speak on YouTube once.)

And the best thing of all (and perhaps even the entire reason for this post, except that it only occurred to me partway through writing it), I can show you the diagrams interactivity (I told you it was an interactive diagram, right?) by making and posting an animated gif. I have a secret fantasy of posting on the Towards Data Science blog, and that site has more animated gifs than a cat has kittens! So I taught myself how to make one to wrap up this article. Enjoy! (Remember to upgrade your scikit learn to 0.23.0 and use those two config lines; then just input the name of your instance of a model and run the notebook cell in order to see its diagram. And see you next time!)