Run your first training session
The results won't be good. That's expected and useful — early failures reveal what needs
to change in your data and your approach before you've invested heavily in either.
Choose a model and set up a shared workspace
Start with the simplest model that could plausibly work. Make sure everyone on the
project has a shared space to log decisions, observations, and questions as they arise.
Apply your domain knowledge
Point the model toward what actually matters. As a scientist, your understanding of the
underlying system is an asset a generic approach doesn't have. Use it.
Days 1-30
Phase 1: Get your first model off the ground.
Collect and label your data
Gather labeled data examples your model will learn from. Quality matters more than
volume at this stage.
Review your data thoroughly
Look for missing values, inconsistencies, and gaps. Set 20% of your data aside now and
don't touch it — this is your test set, and it only works if the model has never seen it.
Define the problem
Write down the decision you're trying to improve in a single specific sentence. If you
can't do that yet, this is where to spend your time before anything else.
26

View this content as a flipbook by clicking here.