Coding and AI classes in Neuss

Source: the tools' own terms of use as collected in our AI tool age guide, checked for the guide and quoted there. The first table is ours. We never ask a child under 13 to open a chatbot account, and we do not use 18+ tools with anyone under 18.

import csv
from collections import Counter

trees = []
with open("neuss_trees.csv", encoding="utf-8") as f:
    for row in csv.DictReader(f):
        size = (float(row["height_m"]), float(row["crown_m"]), float(row["girth_cm"]))
        trees.append((int(row["id"]), row["genus"], size))

train = [t for t in trees if t[0] % 5 != 0]
test = [t for t in trees if t[0] % 5 == 0]
print("Training trees:", len(train), " Test trees:", len(test))

common = Counter(g for _, g, _ in train).most_common(1)[0][0]
right = sum(1 for _, g, _ in test if g == common)
print("Always guess", common + ":", round(100 * right / len(test)), "%")

def guess(size, k=5):
    near = sorted(train, key=lambda t: sum((a - b) ** 2 for a, b in zip(t[2], size)))
    return Counter(g for _, g, _ in near[:k]).most_common(1)[0][0]

right = sum(1 for _, g, size in test if guess(size) == g)
print("Nearest neighbours:", round(100 * right / len(test)), "%")

Sixty-seven per cent against a lazy 39 is a real result, but there is a flaw in how "closest" is measured. The distance adds up differences in height, crown and girth as plain numbers. Height is in metres and differs between trees by a few units; girth is in centimetres and differs by a hundred or more. So a 10 cm difference in girth counts as much as a 10 m difference in height, and the model is really listening to the trunk alone.

The fix is to put the three measurements on the same scale before measuring distance. Version 2 works out how much each measurement usually varies in the training trees (its standard deviation) and divides by it, so that "one unit" means "a typical difference" for all three. It is the same three trees and the same five neighbours; only the ruler changes.

import statistics

spread = [statistics.pstdev(t[2][i] for t in train) for i in range(3)]
print("Spread of height, crown, girth:", [round(s, 1) for s in spread])

def scaled(size):
    return tuple(v / s for v, s in zip(size, spread))

train = [(i, g, scaled(size)) for i, g, size in train]
test = [(i, g, scaled(size)) for i, g, size in test]

right = sum(1 for _, g, size in test if guess(size) == g)
print("Nearest neighbours, scaled:", round(100 * right / len(test)), "%")

Girth varies about fifteen times more than height in these numbers, which is why it drowned out the other two. After scaling, the model gets 70 per cent right. That is a modest gain, and the honest part of the lesson is looking at where it came from, kind by kind, rather than trusting one number.

Data: Stadt Neuss, Baumbestand, licence CC0, downloaded 3 November 2026. The selection, split, model and counts are our own. All three kinds improve with scaling, but the horse chestnut stays hard: size alone gets just over half of them right, a sign that the model needs other information, such as leaf shape, that the register does not hold.

1. Children, 8 to 12. Measure five family members' height in metres and shoe length in centimetres. Who is "closest" to whom if you just subtract the numbers? Does that seem fair?
2. Teenagers. Run Version 1, then remove girth from the size tuple altogether and predict whether the score rises or falls before you run it.
3. Teenagers, further. Change k from 5 to 1 and to 25. Which kind of tree gains, and which loses?
4. Adults. Replace the standard deviation with the range (largest minus smallest). Girth in this file runs from 3 cm to 565 cm; explain why a single extreme trunk can stretch the range and make it a poorer ruler.

We never write a competition project for a learner. We teach the methods, ask the questions a jury would ask, and leave the project in the learner's own hands. The tree model above is the kind of starting point that can grow into one, with better data and a fair test.

Neuss keeps German legal time: UTC+1 in winter and UTC+2 in summer, under the Einheitenzeitgesetz. The last change was on 25 October 2026 and the next is on 28 March 2027. India does not change its clocks and stays on UTC+5:30, four and a half hours ahead of Neuss until March and three and a half after. These are illustrations; no batch is tied to them.

Dates as printed in the KMK holiday overview for the 2026/2027 school year (version of 9 October 2025). Lessons carry on through these breaks unless you choose to pause.