Machine Learning with Nexosis

-

Last night I attended the Ann Arbor .Net Users Group and the topic was "Machine Learning for Fun, Finding Bigfoot". The speaker, Guy Royse, did a really good job presenting the subject. The subject matter made it really entertaining and went beyond just Bigfoot data. There were also discussions of UFO's. It was hard not to laugh as Guy was describing how he obtained the data for the discussion, and how the classification system works. The takeaway from the talk was that this was a fun, lighthearted way to play around with machine learning. 

The Nexosis tool looked really cool. It starts out free, and gives you full functionality until your volume gets to the point where you have to pay for it. It takes quite a bit of volume to get to the paid level, so you can do a lot of development and learning for free. There is a web UI, REST API, and a number of language specific libraries that you can use to access Nexosis. Guy presented several examples using the web UI and the REST API as well as sample code in Python. 

The really cool part of the application is that it does so much of the data science for you. Unlike some of the bigger ML systems, you can upload your data and the application will automatically build your model for you determining the weight of each column, so for a developer who wants to start working with ML, this is a great way to get started. 

I think my favorite part of the discussion was the classification of the Bigfoot data. The data was classified by the type of report and they were similar to the close encounters system for alien encounters except instead of using numbers, they use letters. A Class A sighting is where the person reported actually seeing Bigfoot. A Class B sighting is where the person saw evidence of Bigfoot, and my favorite, a Class C sighting is where someone met someone else who saw Bigfoot. 

After loading the data in the system, it was really simple to create statistical models based on the imported data. In the presentation Guy showed sightings by month and year as well as a distribution on a map. You could make some pretty educated guesses as to why there were more sightings at one time or another. The summer months had more sightings because there are more people in the woods. The UFO data was even more interesting because the largest number of sightings of UFO's happens to fall on the 4th of July. Obviously our alien visitors love stopping by to watch the fireworks. They probably get a pretty good view from their spaceship.

After setting up the model, Guy also showed how you could pass text of a new sighting to the system, and based on language recognition, the Class of the sighting could be determined. He showed a couple of examples that worked really well, and then one that didn't and explained how the text confused the system. 

In the end, we didn't actually find Bigfoot, but we got some pretty good insights into the data. If you get a chance to see Guy Royse talk, I would highly recommend it. He is going to be at Beer City Code, https://beercitycode.com, this weekend, and you can find out more about him on his blog at http://guyroyse.com. If you would like to learn more about Nexosis, their website is: https://www.nexosis.com.

If you are interested in the Ann Arbor Dot Net Users group, their website is https://www.meetup.com/aadndevs Next months talk is on Defending Against Advanced Web Application Threats.

Do you have any thoughts on Machine Learning, Bigfoot, or UFO's? Let me know in the comments. 

Comments

Post a Comment

Popular posts from this blog

Asp.Net Core with Extended Identity and Jwt Auth Walkthrough

-
Image
  Quite often I think of an idea for an app where I would have a website and a mobile app that share the same functionality. Every time I start to prototype something like this I get stuck on authentication. When you build a web application in Asp.Net Core, it is really easy to set up the built-in Identity platform to allow users to login, but if you want to allow a mobile app to access APIs on the same site, that takes a lot more work. Of course, you can use a cloud-based authentication system like AzureAd or Firebase authentication, but if you arent housing any sensitive data, that's a lot of extra plumbing that takes away from building your application.  This walkthrough will show you how to create an Asp.Net core site with Asp.Net Core Identity for authentication and authorization. We will be extending the identity to include additional user information and roles. We will also add jwt authentication and authorization, so SPA's and mobile apps can use the API's and share...
Read more

File Backups to Dropbox with PowerShell

-
Image
I have had a number of instances where I had a virtual server somewhere, and I wanted to have a backup strategy, but the paid plans were overkill for what I needed. I have all my source code in Github, so everything can be rebuilt. I just need to be able to archive database backups and uploaded files. I decided to use a Dropbox account and set up a scheduled task that calls a PowerShell script that accesses the Dropbox web API. Setting up Dropbox You will need to have a Dropbox subscription set up and then you can go to the developer's section at the following link: {% embed url=" https://www.dropbox.com/developers " %} On the main page, select create apps. On the following page, you will have the option to select the scope of the app. Here you can decide to only access a single folder or give access to all of the Dropbox functions. Either option will work, in my example, I used the full access scope, so I would have the option of putting different file types in different...
Read more

Dynamic Expression Builder with EF Core

-
I was recently working on an ASP.NET Core MVC project where the end user wanted to have a search form with several fields with drop down boxes and a couple of free-form text fields. One of the requirements was to be able to put "*" at the front or the end of a search term in the text boxes to indicate a wildcard search. That was the way a previous application worked and they wanted to maintain the same functionality. Since I was working with Entity Framework Core using a generic repository pattern, I wanted to be able to dynamically build the filter expression that I passed to the repository. I created a static Expression builder class with static methods for creating each type of expression needed. public static class ExpressionBuilder { public static Expression GetStringExpression ( string dataModelProperty , string dtoProperty , Expression combinedExpression , ParameterExpression parameterExpression ) { if ( dtoProperty...
Read more