Planning future experiments

Now that you have an idea of what model size you’d like to target given your compute budget and data constraints, let’s learn how to think about each run of your job as an experiment. Fundamentally, each stage in the machine learning process is ultimately a unique experiment. Some of your inputs to each stage in the project stay the same; you could call these your dependent variables. Some of your inputs to the project change; these are your independent variables. it takes time to build up skills on your project Simply put, change something and see what happens. Just make sure you’re only changing one thing, so it’s empirically clear what the result is!

It’s critical to understand that the whole scope of your project is not going to happen all at once. A lot of this is because it takes time to build up skills on your time. Even if you are starting with a completely experienced team, which frankly happens very rarely...