This lists will keep being updated by suggestions of friends & colleagues, or whenever I feel enlightened.
Tips
-
Before you run your code, have clear expectations of what this code will do.
This will help you build a solid understanding of what you are building, why stuff works and, most importantly, why stuff don’t work. Additionally, splitting the code into small but clear-cut functions / classes can help you both maintain faster and debug easier by narrowing down the issue to specific modules.
-
Make short term goals.
Such as a weekly checklist. And don’t delete it at the end of the week! Saving in notes the progress you went through during the previous weeks can help your morale and motivate you when you are questioning the value of your work.
-
Iterating too slowly.
You will. There is no question about it. It might be for a week, it might be for a month. At some point, you might reach a plateau because of x and y, that’s fine. However, if you plan for this, you can leverage this “not-so-productive” time, to make progress on other tasks, like writing your report, catching up on related work, refactor or optimize old code, cleanup the repo, make a nice readme etc.
-
Keep notes of every (yes, every) run you make.
Why you made it, what were the expected results and what were your actual results. After working in a project for a long time and something starts going wrong after months of work, you can lose your sanity (and time) by questioning every part of the code from the start. If you have a thorough straightforward notebook of experiments / tests you have done in the past, it can help your future self understand why stuff are they way they are.
Tools I could not imagine doing my thesis without
at least +25% productivity gain
- Overleaf (Online LaTeX Editor)
- GitHub (Version Control)
- Weights&Biases (Model tracking)
- but you could also use: mlflow, neptune.ai, guild.ai, comet.ml
Tools that helped
- Zotero (Citation tracking)