It's dissertation time

And here we are, at the pointy end of my part time master's degree! If I'm really good, I'll be able to finish it within 3 months. If not, I get the luxury of up to a year, but honestly I don't fancy dragging it out for so long.

Before shedding a bit more light onto my dissertation, I thought I'd reflect on the second year of studying first. And what a year it was! I make no secret that I'm aiming to scrape a distinction, and to do so, I need to score an average of 60 in the taught modules and 70 in the dissertation. I safely navigated the first part, with some room to spare, so I'm really pleased about that.

Year 2 was building on what we had learned in year 1. I also decided to go down the technical pathway instead of general pathway, and ended up with the big challenge of trying to understand reinforcement learning. This was a great module actually, and I learned a lot about how games are played and how many of the python libraries like Keras and Tensorflow are used. There was a group assignment too, where we had to work on an appropriate project, so we chose the Lunar Lander Atari game. It was a lot of fun, but tough for sure.

There was also a module about robotics, but it was sadly more theoretical than practical. The assignments in this module were both essay based, and while I learned a lot about researching and the subject matter, it felt like it was a bit of a waste not to actually work on anything to do with making robots work. I feel now though, that I'm an expert on how my robot vacuum cleaner uses the SLAM algorithms to map my home.

The other major highlight in year 2 was the Further AI module, where we had to make both a decision tree and genetic algorithm from scratch. This was really good, and I felt like I learned a lot about how both of these types of algorithms work. Learning how to put them together, seeking out the information and tutorials were great fun, and piecing it all together gave me a huge sense of satisfaction.

I also noticed that this year included a lot on analysis and reporting of our coding. This was a shift in what we had done in the first year, and I really relished this. It tests a different side, and I do feel that it has given me a chance to demonstrate my critical thinking and analytical ability. It's reflected in the results, and I'm very proud of how I've performed overall, especially as I thought I'd crumble this year. I have had one or two duff results, of course, but so far I'm still comfortably in distinction territory, and that's a good thing.

The final module of the year involved preparing for the dissertation. This was a real challenge, and the way this module worked just showed how independent we would need to be. It was not easy at all, but I drafted up and designed my dissertation around a subject I know well: Google Analytics. The overall gist of my dissertation is to try to build a methodology that can be used to define user personas based on Google Analytics data. Secondary to that, those persona parameters should then be used for machine learning application of predicting the type of user within any given session.

This perhaps doesn't sound particularly ground breaking, especially as many marketing tech companies and ad networks all push their own machine learning. But why should only those companies have the secret all to themselves? What about all of the website owners out there who could do with maximising their marketing budgets and understanding their website visitors? I wouldn't exactly say that I'm exposing the industry, but those in the industry know that much of this is secret work, and none of it is particularly open for sharing.

I'm about 2 months into my work, and I'm not sure if I'll be able to make the 3 month deadline. In all likelihood, it'll take me 6 months to complete, but who knows, maybe I'll be able to blast through in this final month. I've written my background, introduction, literature review and methodology, so I'll need to go back and update and proof read. I'm in the process of experimenting and writing up my work, so hopefully I can get through that soon too. I just wish that the university support in my case was a lot better.

Overall though, I'm proud of the work I've done so far - even with basic algorithms and tuning, I've managed to find a way that takes almost half a million rows of data, selecting the most appropriate columns, and then building a machine learning predictive model that is approximately 92% accurate. That's definitely not bad in my book!

I'd better get back to the work. Because of the size of my dataset, running the code takes some time, so I must find a way to be productive while I'm doing more tuning. Doing so, means that I'll stand a chance of meeting that 3 month deadline, and then I can reclaim my nights in front of the TV once more!