Hackathon. Check.

Last week, Ayasdi hosted its first ever internal Hackathon. We are a pretty geeky crew over here and we have sponsored our fair share of Meetups, but we have never done a Hackathon – so we were overdue. It seems that we had some pent up demand though – because teams across the company responded with exceptionally cool stuff.

While the teams indicated they would do it for fun, truth is there was money involved. Not, “buy a new car” money, but take your significant other to The Village Pub (hint: it’s not a pub) kind of money…

Ultimately the company self-assembled into 8 teams competing for four prize spots (a winner and three runners up). Here is a list of the winning projects:

    1. Under the Olympic theme of Better, Stronger, Faster, the team of Rayan Seshadri, Anhad Bhasin and Shidong Wang embedded Spark capabilities into Ayasdi’s Machine Intelligence Platform to deliver distributed compute scale and performance gains. What kind of gains? More than 40x faster with an accompanying 8x reduction in memory consumption over workloads of 10M rows and 700 columns. Pretty good start. The best part about it? It should be in production by the next release.

    2. Predicting Traffic Accidents – Perhaps the least appreciated feature in our software is our Topological Predict functionality. Our experience is that most problems are a combination of something unsupervised and something supervised. Segmentation + Prediction, Anomaly Detection + Prediction etc. Bill Oliver was a one man wrecking crew in the hackathon and extended our functionality in this area. His example was to combine date/time, lat/long and other data to predict traffic accidents – both by location and by time. Bill tackled the challenge that is inherent in geo-coordinates, that latitude-longitude are, by themselves, meaningless, so the must be used together and only together. While Bill’s work looked at traffic, it could easily be applied to tornados, terrorist attacks and other phenomena.

    3. Ayasdi’s Machine Intelligence Platform on Mac – We are pretty much a Mac shop over here with a handful of exceptions (sales…). Yet for some reason, we didn’t support Ayasdi’s Machine Intelligence Platform running natively on a Mac. Well, the team of Ryan Hsu and Samuel Delano fixed that in record time. While this is not something our customers will see, it is something they will feel. The reason is that it enables our developers to much more rapidly develop code, test it, and get to the root cause of a bug.  It should substantially improve our development cycle.

      Very cool stuff.

    4. Novel TDA Visualization: 3D and VR visualization of TDA networks – It is no secret that our topological models are gorgeous. The way they elegantly encode information is remarkable and accelerates the discovery process for our clients. The team of Conor Hanrahan, Pankaj Dubey, Bob Pappas thought it would be interesting to go inside the network and to see it from different angles. Ultimately it required the team to render our networks in 3Dwith a coloring as the 3rd dimension. X, Y, force, a coloring for node color, and another coloring for Z.   By taking this approach, they were able to visualize a 4th dimension of the data x,y, color and height(z). For example coloring the network by variable direct cost and then using hight to represent  the frequency of visits to surgery center. Surprisingly, this immediately showed that cost was not correlated to visits to the surgery center.

      Trimmed-Complex-Network

 

While the winners get the credit there were a number of other “honorable mention” projects that covered a typically wide range, from extracting text from Reddit to predict changes in crypto prices, to credit card fraud detection to annotation widgets for our Workbench product.

 

Ultimately though, a hackathon serves a purpose, to get us out of our tracks and to indulge our creativity. We are blessed with an exceptionally technical team at Ayasdi and it was amazing to see what they could create in such a short window. Keep an eye out for some of these features in future releases and demos!