Awesome F# - Decision Trees – Part I

re: Awesome F# - Decision Trees – Part I

ChrSmith — Tue, 03 Nov 2009 18:12:24 GMT

Re: multiple regression analysis, if there is a clear trend in the data you can likely arrive at it through multiple means. (In the Evil Genius example, having a Ph.D. is the only significant bit.)

However, decision trees offer a few advantages. For example, there are some algorithms which work 'online' meaning that they build up the decision tree as training data comes in. (Meaning it gets more accurate over time and doesn't need 'learning time'.)

There are many things not covered by this blog post such as overfitting your training data or boosting to improve your trees accuracy. If you want to learn more about techniques for machine learning Tom Mitchell's book is really, really good.

re: Awesome F# - Decision Trees – Part I

Hank Fay — Tue, 03 Nov 2009 15:05:44 GMT

Hmm... I look at that data and I see a multiple regression analysis (which definitely does not require F#) of the sort used by Bottenberg & Ward for fight pilot selection in the early 60's (yes, I am that old), in which the contribution of each variable is assessed, individually and in combination (the same analysis can be done using factor analysis as you might guess). Of course one would like more data, but the error term will be the same no matter what process is used for analysis. Would the results be the same, do you think?

re: Awesome F# - Decision Trees – Part I

A Fan! — Mon, 02 Nov 2009 07:59:28 GMT

Please more like this (i.e., applications or how to use F# in to solve problems)

re: Awesome F# - Decision Trees – Part I

ChrSmith — Sun, 01 Nov 2009 17:08:34 GMT

Yes, machine learning definitely falls into the "it's easier to do in F# than C#" camp due to the mathematical nature. I'm happy to hear you're putting F# to good use.

re: Awesome F# - Decision Trees – Part I

Jim Martin — Sun, 01 Nov 2009 16:06:07 GMT

Awesome job Chris! That is the same book we are using in my Machine Learning class, it's a great book. Can't wait to see your final solution. I got mine finished and turned in and got a 100% on it. It was a lot of fun and challenging at the same time.

A lot of people in the class were doing this problem in either C# or Java. I asked the prof if I could use F# and he said sure. I had to learn F# as I did the problem so it was quite challenging. I've since used F# to model on Neural Nets and Perceptrons, and today I'm working on a problem around an MDP. F# has been awesome for these Machine Learning problems.