2

I have a few months to go in my MASc program in Aerospace Engineering. I'm not tight for time or anything, but I'm curious if anyone has tips on how to manage results.

My research is highly algorithm-centred, and there are a lot of little knobs and dials on the algorithm that I added as I went along, and I'm still tinkering with it. I know what they all do, but there's no way I can do a detailed analysis on the effect of each setting.

My advisor says it's important to have the results in say figures and charts etc exactly reflect the algorithm as written, which I agree with. But how will I know when the algorithm won't change anymore, that this is how it will be, and these are the final results? Otherwise I'll end up creating the figures 10x over. I could leave the figures out till the very end, but often the discussion is around features in the figures.

I know there's no silver bullet, but wondered if anyone had good tips on managing that sort of problem.

1
  • 4
    Be careful that your results aren't too dependent or overfit on the dials and knobs you keep adding :)
    – Bryan Krause
    Jan 29 at 2:44

2 Answers 2

5

When I try to present my results, either by text or even more by graphs, plots, figures, I often find questions or issues I have to investigate. So not writing a draft and creating the graphs, stresses me after I finally do so.

To address your problem with doing graphs over and over, you have to automate its generation. I wrote all of my theses with LaTeX so TikZ and its plotting functions were an obvious choice. But you can do the same with matplotlib (Python), Matlab, R, Excel, LibreOffice Calc, Chart.js (JavaScript): Write scripts that read input files or where you have to copy/paste your data. The plot is automatically generated, including postprocessing, scaling, labeling etc.

It is some work, but it pays off! And it adds transparency, because you can always proof that the graphs were not manipulated - or it would be obvious because it is written in your code. Store the code together with the raw data - just in case.

After the final review from my professor, I was asked to drop one data set. I could do this within a couple of hours and it was not stressfull, as 80% of my graphs were automated.

6

I had a similar algorithm-centred thesis for my MASc. To keep the results manageable, I had to pick a certain easily-justified position for all knobs and dials to be my "baseline" setup. This let me still explore the algorithm design space as the thesis due date got closer, but treating each new result as a special case with a short associated discussion. If you can, it may be useful for your reader and easier for you if each knob and dial can be discussed in its own section. If there's stuff you could have explored in more detail, it's normal to acknowledge that at the end. A hard time limit for master's research is always going to cut you short.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .