I started a master’s program in mathematical sciences at Gothenburg University in September 2018 and am specializing in mathematical statistics. The program is very flexible, with the only required classes for the statistics specialization being (measure theoretic) probability theory, as well as a choices between the three courses
- Computational Methods for Bayesian Statistics
- Statistical Learning for Big Data
- Statistical inference principles
As a part of the mathematical science program, I was required to a take a course on perspectives in mathematics (history).
After completing my first year I wanted to review how the program is going so far as well as my goals for next year. I recorded the amount of time I spent outside of my courses using toggl. The study periods are about 9 weeks with a four hour exam for each course.
Time Tracking Analysis
I kept track of my time for each course using toggl.com. All of the time spent is on studying, homework, projects and all else outside of the lectures. At Gothenburg university and Chalmers the year is divide into two semesters HT and VT (fall and spring) and each semester has two parts. I use another app, forest to study in shorter increments, generally in 25 minute blocks followed by a short 1 minute break and then longer breaks after multiple blocks. This method is largely based on the Pomodoro Technique from the popular learning how to learn course on coursera.
We can see in Figure 1, that I spent the most hours on the second part of the spring semester.
I got into a better method of studying in spring and felt much more efficient. The biggest change was attempting to spend a normal work day at school, approximately 8 to 5. Also, the sun was out much longer which definitely helped with my motivation.
What I can improve on
- Spreading out the studying better across the quarter
- Fix: Plan the week out and set mini-goals for each class.
- Handling the gray and dark during the winter
- Fix: I’ll be escaping this winter back to California so I think this will be more manageable. Also collaborating and working in the library around other people helps.
- Documenting what I have done
- Fix: Add interesting topics as blog posts here.
- Courses without assignments or projects
- Fix: Either don’t take courses without projects or do my own independent project.
What I did well
- I put a lot of time into the classes that had projects and assignments.
- Studied what I thought was useful and interesting, rather than just what I needed for the grade.
- Found holes in my knowledge and embraced not knowing things.
Going into next year I want to keep up the regular working hours I had in the second half of the spring semester. Keeping a good routine going will allow me to actually relax on the weekend instead of worrying if I need to be working on something.
It is fairly easy to see from Figure which courses in enjoyed the most.
Statistical learning was a very interesting course where we presented four topics throughout the quarter. The topics were classification, clustering, penalisation/variable selection, and dimensionality reduction/matrix factorization. The subtopics I selected were
- Reduced-rank and regularized discriminant analysis
- Non-parametric kernel-based methods for clustering
- Mislabelling in the context of lasso
- Missing value imputation using matrix factorization
Overall this format worked very well and I learned a lot about these specific topics.
Financial time series was an interesting course for me. I found the topic very dry for most of the semester but towards the end really started to enjoy it. We had two smaller projects, both of which were in matlab. I have no interest in matlab but I did feel like I learned a lot from these projects. The projects were
- Modelling the absolute returns of the Australian Trade Weighted Index
- Check for stationarity
- Ljung-box for iid
- Wrote a best linear predictor
- Tests for normality
- Modelling the log returns of the S&P Europe 350 market index.
- Compared various ARCH/GARCH models using BIC/AIC
- Evaluated using a training and test set
- Compared against a GJR model to model positive and negative volatility differently
I’ll either expand on them in some other post or make an edit here.