This notebook is more of a blog-style where I'll be discussing what I found useful to know and understand about how research is different and ideas/concepts you should keep in mind:
Understanding the Data
- Understanding the ins and outs of the data plays a large role in being able to interpret behaviors about your model and its performance
- Without this we're just throwing a model at data and while we could perform well, you'll be asked why and understanding both the data and the technique let's you do so
Understanding the Techniques
- You'll be the expert in neural networks on the team (I've found), so you need to make sure what you're telling your team is accurate and reliable information
- I found this a particular struggle
- Test Sets
- How (and why) to create a good validation set
Understanding how to Interpret the Techniques
- This is especially hard with tabular, as many researchers still view neural networks as black boxes where we can never know what's going on
- Understanding how to use the various tools to interpret models for your peers/partners in a way that they can understand is extremely valuable. While we may understand what ClassConfusion, SHAP, etc is doing in the background, it's another set of skills entirely to communicate this effectively
You are keeping yourself accountable
If you're ever iffy about anything, automatically assume you need to read a bit more into how to make something work
I had this issue with my research project
- Probably the hardest thing to do is admit when you are wrong, but it's especially important to do so before it's too late