How projects actually proceed.
WHAT THE STAGES ARE
Framing the problem Establishing whether data exists Building a baseline Preparing data and features Training and tuning Evaluating honestly Deploying Monitoring Retraining and improving
WHAT PROPORTION OF EFFORT MODELLING IS
Small. Data work and deployment dominate.
WHAT NEWCOMERS EXPECT
The opposite.
WHAT A BASELINE IS
The simplest approach that could work.
WHAT IT MIGHT BE
Predicting the most common class Predicting the previous value A simple rule the business already uses
WHY IT IS ESSENTIAL
It establishes whether a model adds anything, and many do not.
WHAT TO DO BEFORE TRAINING ANYTHING
Build it, and measure it.
WHAT THE ITERATION LOOP IS
Change one thing, evaluate, keep or discard.
WHY ONE THING
Changing several makes the cause of any improvement unknowable.
WHAT ENDS MOST PROJECTS
Not model quality, but the absence of a path to production.
WHAT TO ESTABLISH AT THE START
How a prediction would actually be used, and by what system.
