It’s been a year since the first blog post about the TMCS hackathon 2017. This year, Dave, Mike, Lars and I went back to Oxford to teach the same software development course for the Theory and Modelling in the Chemical Sciences (TMCS) master students.
The course runs over 4 days, where the first 2 are focused on teaching the basics of python, object oriented programming, IDEs, git, etc. The material for this course can be found on github. In the last 2 days, the students are split over 3 groups and have to work on a small project. All three projects this year involved neural networks, since they are a hot topic in our research group.
My group focused again on fitting a potential energy surface for a cyano radical reacting with a methane molecule. However, this year the data was slightly different. They had about 17000 data points (compared to about 8000 last year) and the data was calculated at CCSD(T) level.
I also made another change to the project. I created in advance a file for each function that the students had to code, and specified what shape and type the input and output of each function should be. This was to minimise problems when all of the functionalities had to be brought together. Therefore, this iteration of the Hackathon was smoother than the previous one, although I will have to make a few more adjustments for next year.
At the end of the 2nd day, a neural network was fit to some data and some correlation plots were made! No time was left to do any hyper-parameter optimisation for the network, so the prediction errors were still quite high. However, it was a great success, as everyone finished their assigned tasks and all the separate pieces of code worked well together. So, well done to Hannah, Isabel, Ben, Jamie and Harry from the 2017 cohort!