Tech Blog

Jay's Technical blog

Machine Learning-Types

04 April 2016
Jay Kimble

o this is a quick one off. I’m actually working through the MS Virtual Academy Machine Learning course (and will probably go after the certificate). One of the things I am still have problems remembering is the various (common) types of Machine Learning. These things actually build an algorithm to resolve a particular type of problem.

So this is a quick one off. I’m actually working through the MS Virtual Academy Machine Learning course (and will probably go after the certificate). One of the things I am still have problems remembering is the various (common) types of Machine Learning. These things actually build an algorithm to resolve a particular type of problem. It’s pretty cool because you provide it with a set of data and it does something with that data. So without further adieu, here is my notes on these types:

Classification: This ultimately answers a yes/no question or true/false assertion. So if you want to predict the answer to a yes/no question. For instance, you might want to know if the Cleveland Browns win the super bowl this year; Classification will predict this (if you need ML to learn this, then you haven’t been paying attention.. the answer in Spring of 2016 is a definite NO).

Regression: This is similar to the last one except you want to predict some outcome. For instance, how many points will the Cleveland Browns score this year (you are probably seeing a pattern here)? BTW, this number will probably be a low number.. possibly less than 100..

Clustering: This groups data together. Using different types of clustering will net different types of results. Let’s say you plugged in NFL football teams and put in there record each year over say the last 15 years as well as the number of points scored, super bowls won, etc. You could then have it group these teams into groups. You can tell it how many groupings you want, etc. (of course the Browns would be grouping of losing teams). You could also plug in team colors, and the team logo as opposed to win loss stats, and the machine learning might group teams by color and maybe even by whether the mascot is an animal or something else. (Clustering can interpret images)

Recommender Systems: Historically this is the one we know the most about. With e-commerce systems we’ve wanted to predict what other products a user might want to buy. This is exactly what this type of machine learning does. (another question might be, “after following the Browns for 35 year, which NFL team should I switch to that feels the most like my team (but with a winning record?” Ok, maybe I’m not ready for that question).

  ML

Machine Learning-Journey

04 April 2016
Jay Kimble

I’ve really been interested in systems where the computer could determine what the user was doing and drill them into a better workflow (or alert the programmers that we need to build a better workflow). Or, even things where what the user is doing will ultimately fail (maybe they are generating a contract that will fail..

So, last year, I spent some time doing everything using JS. I have recently begun a new journey around Data Science and specifically Machine Learning.

Now I’ve primarily been a front end dev, and in my mind that means that I tend to do a lot of work in JavaScript and HTML, as well as C# and XAML. This has been where I have been most skilled in the past. Or at least what you have seen in my recent writings.

I’m also a very good Database guy. I brag that I’m not actually a DBA but in a pinch I can stand in the place of one. What I mean by that is that I can write SQL (queries, procs, functions, etc.) I have a decent head for troubleshooting here as well as a decent head for performance analysis (I’ve rarely seen another dev pull up the query perf analysis in SQL Management Studio). A long time ago I worked on Oracle (and I still have some of those Performance skills). I’ve done PostrgreSQL and SQLite, etc. I’ve also done some lightweight playing around with MySQL, and have played around with NOSQL database (mostly MongoDB.. I can say MongoDB is really interesting to me). Anyway, the point is that I’m also a data guy.

When it comes to bringing the two together, I start to lag. I hate writing service tiers. I just don’t find them interesting. I have friends who love this space because they are predictable (I think that’s what takes the fun out of them).

Anyway, I owe my interest in Machine Learning to Shawn Cady who was my boss at another job. A couple of us ran a UX user group (which was infamous), and one night Shawn started talking about an article he had read about the potential on UX using Machine Learning (although at the time it didn’t have this name; we thought more about recommendation systems).

I’ve really been interested in systems where the computer could determine what the user was doing and drill them into a better workflow (or alert the programmers that we need to build a better workflow). Or, even things where what the user is doing will ultimately fail (maybe they are generating a contract that will fail to make money or will fail to be approved, etc). These types of systems are really useful because they save the user time as well as help us programmers build better systems.

I will try to do a better job of documenting this journey (I really do want to blog more).

  ML