Introduction
Imagine you have a set of data that consists of two types of points. We will call them type A and type B. When you plot out the points type A points are on one side and type B are on the other side, so to divide them it’s simple enough just draw a line between the two sides. But what if it wasn’t that easy, What if when you plotted them some of the type A points were on the B side and vice versa? Well, this is where you would use a support vector classifier. Well, that is what a video by Cassie Kozyrkov in her course Making Friends with Machine Learning discusses(link to video found here).
Summary of Article
The video starts out by discussing what support vectors are. These are the “points that matter” when you are classifying data in this way(drawing a line between the two types of points to classify them). They are the points that are close to the line that you draw. If one of those points moves a little bit you are much more likely to need to change the line vs. if a point well within its section moves a little bit. The video then gets into the scenario I depicted above where some of the type A points are on the type B side and vice versa. Here you would use a support classifier, and what this would do is penalize the points that were on the opposite side and still draw the line in between the two sides. This is better because it allows you to keep the general trend of the data while taking into consideration that there are some points of type A that are on the type B side and vice versa. This reduces the probability of overfitting to your specific sample of the overall population and increases the chances that your model can be applied to more data and still hold its accuracy.
My Take
Overall, I really liked this article. It helped explain a very complicated-sounding concept in layman’s terms. I also really liked how Cassie started out by explaining what a support vector is and then moved into support vector classifiers. That step in explaining the foundation and then moving on to the more advanced concept really helped me understand these concepts. I also really liked Cassie’s visuals in her talk because they helped me understand these classifiers and how they draw these lines to separate data into groups. One thing I didn’t like about the video was how it started in what felt like a very abrupt spot, and the audience didn’t feel like they had all the context they needed to understand the slides and what Cassie was talking about so I wish that the video would have started earlier to understand the scenario being described in the first parts of the video.
Conclusion
All in all, this was a great article. It helped me understand complex concepts easily, and the visualizations were really good in aiding with that learning. I wish that more context was given at the beginning of the video. But overall, it is a really great video and I highly recommend you check it out(link to video found here).