Write a real machine-learning model yourself: to guess a new penguin's species, find the most similar penguin in the data and copy its answer.
Music apps suggest songs like the ones you play; shops suggest things bought by people like you. Many of these start from one idea: find the closest match.
Instead of writing rules, let the data decide. To guess a mystery penguin's species, find the penguin in the data whose measurements are closest, and give its species. This is called nearest neighbour, and it is a real machine-learning method. Closeness is a distance: with bill and flipper length as the two sides of a right-angled triangle, Pythagoras gives the distance between two penguins (** 2 squares a number, ** 0.5 takes the square root).
Make your prediction first. Help opens once you've locked it in.
def distance(a, b):
return ((a["bill"] - b["bill"]) ** 2 + (a["flipper"] - b["flipper"]) ** 2) ** 0.5
print(distance({"bill": 40, "flipper": 190}, {"bill": 43, "flipper": 194}))
The bills differ by 3 and the flippers by 4. What does Pythagoras say?
Run it, then change the mystery penguin's measurements and run again: try 38, 185, 50, 196 and 45, 205. Which is the hardest to guess, and why?
import csv
penguins = []
with open("penguins.csv") as f:
for row in csv.DictReader(f):
penguins.append({"species": row["species"], "bill": float(row["bill_length_mm"]), "flipper": float(row["flipper_length_mm"])})
def distance(a, b):
return ((a["bill"] - b["bill"]) ** 2 + (a["flipper"] - b["flipper"]) ** 2) ** 0.5
def nearest(new):
best = penguins[0]
for p in penguins:
if distance(p, new) < distance(best, new):
best = p
return best
mystery = {"bill": 47.0, "flipper": 215.0}
match = nearest(mystery)
print("Nearest penguin: " + match["species"] + ", bill " + str(match["bill"]) + ", flipper " + str(match["flipper"]))
print("Distance: " + str(round(distance(match, mystery), 1)))
print("Guess: " + match["species"])
import csv
penguins = []
with open("penguins.csv") as f:
for row in csv.DictReader(f):
penguins.append({"species": row["species"], "bill": float(row["bill_length_mm"]), "flipper": float(row["flipper_length_mm"])})
def distance(a, b):
return ((a["bill"] - b["bill"]) ** 2 + (a["flipper"] - b["flipper"]) ** 2) ** 0.5
def nearest(new):
best = penguins[0]
for p in penguins:
if distance(p, new) < distance(best, new):
best = p
return best
mystery = {"bill": 47.0, "flipper": 215.0}
match = nearest(mystery)
print("Nearest penguin: " + match["species"] + ", bill " + str(match["bill"]) + ", flipper " + str(match["flipper"]))
print("Distance: " + str(round(distance(match, mystery), 1)))
print("Guess: " + match["species"])
Answers from examples.
What the distance says.
The furthest penguin, not the nearest.
The mystery penguin (47, 215) is guessed as an Adelie 44 away, though a Gentoo 0.2 away is in the data. Click the line that is wrong.
Ask for the bill and flipper length with input().
Replace the mystery line: bill = float(input("Bill length (mm)? ")), the same for flipper, then mystery = {"bill": bill, "flipper": flipper}.
When the nearest penguin is more than 10 away, print Not sure instead of a guess.
If distance(match, mystery) > 10, print Not sure instead of the guess.
Ask for the mystery penguin's measurements with input().
Run your program. This ticks when your code does all of the above.
Print Not sure when the nearest penguin is more than 10 away.
Run your program. This ticks when your code does all of the above.
Guess the species of each of the five mystery penguins, and print each guess with its distance. Which guess would you not trust?
Guess the species of each mystery penguin, and print each guess with its distance.
Run your program. This ticks when your code does all of the above.
How good is this model really? Testing it on penguins it already knows would be cheating: that is the next lesson.
Students write a nearest-neighbour model from scratch, measuring the distance between penguins with Pythagoras and copying the species of the closest one, then ask for measurements, add a "not sure" answer and guess five mystery penguins. About 45 minutes.
Open the teacher guide: answers, solutions, slides and a worksheet.
Read the unit primer first. Nearest neighbour keeps every example and answers with the species of the closest one; the distance is Pythagoras on bill and flipper length (** 2 squares, ** 0.5 is the square root). nearest() is Data detective's "smallest so far" pattern (the bug task uses >, finding the furthest penguin). A big distance means nothing similar is known, so the guess is weak, but the model still answers: the "Not sure" task and the optional last lesson come back to this. The info question raises scale: flippers (around 200) outweigh bills (around 45) in the distance.
The smallest-so-far loop inside nearest(). Ask students to trace it for three penguins on paper before they fix the bug.
Stretch: Make the three nearest penguins vote (sort by distance with sorted(penguins, key=lambda p: distance(p, mystery))[:3]).
Support: Give students the Not sure if, and ask them only to choose the number.
Next lesson tests the model fairly, on penguins it has not seen. scikit-learn's KNeighborsClassifier (Lesson 4) is this model.