Exercise 3: What Step 7 Does vs. What ml1 Does — Possible Solution ==================================================================== WHAT STEP 7 ACTUALLY PRODUCES, PER THIS CHAPTER ------------------------------ Per this chapter, Step 7 is "Form hypotheses... the actual point of everything above — turning patterns into specific, testable questions." The chapter's own worked examples are phrased explicitly as QUESTIONS: "Does mileage predict price more strongly than the car's year does?" "Is the Jaguar a genuinely different category of listing?" "Does fuel type meaningfully shift price once mileage and year are accounted for?" Every one of these is a question awaiting an answer, not a conclusion or a working predictive tool. WHAT THE WARN-BOX SAYS ml1 ACTUALLY DOES ------------------------------ Per this chapter's own warn-box, "none of these questions get answered by anything in this chapter. EDA's own job ends at forming a specific, well-motivated question backed by a real pattern already seen in the data — actually building something that predicts price from mileage and year is ml1's entire job, start to finish." ml1's own job is constructing an actual model — a system that takes mileage and year as input and produces a predicted price as output, trained and validated on real data — a fundamentally different kind of deliverable from a well-phrased question. WHY "DOES MILEAGE PREDICT PRICE?" IS RAISED BUT NOT ANSWERED HERE ------------------------------ This chapter's own heatmap (Step 5) can show that mileage and price correlate — a real, computed number confirming SOME relationship exists — but per ds1-6's own correlation-vs-causation material, correlation alone was never enough to establish that one variable reliably PREDICTS another in a usable, quantified way, let alone build a system that outputs an actual price estimate from a given mileage value. The heatmap only confirms the QUESTION is worth asking — it doesn't build, test, or validate a predictive model that could actually answer it with a specific price estimate for a specific new car. WHY THIS DIVISION OF LABOR MATTERS ------------------------------ Per the warn-box, "this chapter's own seven steps are the bridge that decides what's worth building in the first place, not the construction itself." Without EDA, ml1 would have to guess blindly at which variables might be worth modeling; without ml1, EDA's own well-formed questions would simply remain unanswered forever. The two are deliberately sequential and complementary, not overlapping — exactly matching the workflow ds1-1 first laid out (Collect → Clean → Explore → Model → Communicate), with this entire course stopping precisely at the Explore/Model boundary. WHY THIS WORKS AS AN ANSWER ------------------------------ It contrasts Step 7's actual output (a specific, testable question, per the chapter's own three worked examples) against ml1's actual output (a working predictive model, per the warn-box), and explains specifically why a correlation heatmap can motivate a hypothesis without being capable of answering it, since confirming a relationship exists is a different task from building a system that predicts from it.