Shopping Segments

Part of the Unsupervised Learning section of Coddy's Machine Learning journey. Lesson 48 of 51.

An online plant shop knows its 240 customers only by what they do: orders a year, basket size, share bought on sale and days since the last order.

The plant shop's 240 customers as gray dots, orders per year across and average basket in euros up, with no labels: a dense band at low baskets and a small high-spending group near the top left

No one has labelled them. The owner wants a few segments she can describe in plain words, and a win-back email for the one that drifted away.

Let two methods agree on k: a ward tree suggests it, and the silhouette checks it against its neighbors. Then read every segment in real units.

Challenge

Hard

The program reads where to cut the tree; the starter sets up a titled chart.

Scale the features and cut a ward tree there. Print how many segments it makes.
Print the k-means silhouette (3 decimals) for that k and each neighbor, then the best k.
Refit with it into plants["segment"]. Biggest first, letter each segment and print its real-unit means; then the one whose last order is oldest.
Scatter each segment labelled by its letter; legend, show.

For another shop, cut at 8:

Tree cut at 8.0: 3 segments
Silhouette at k = 2: 0.573
Silhouette at k = 3: 0.675
Silhouette at k = 4: 0.630
Best k: 3
A (70): 8.6 orders a year, 29 EUR baskets, 15% on sale, last order 23 days ago
B (50): 3.1 orders a year, 22 EUR baskets, 60% on sale, last order 53 days ago
C (30): 1.8 orders a year, 89 EUR baskets, 10% on sale, last order 148 days ago
Win-back email: segment C

and its chart:

Another shop's segments titled Plant Shop Segments: A, 70 customers with 5 to 14 orders a year, B, 50 with few orders and small baskets, C, 30 with few orders and big baskets near the top, with a legend

Then Coddy describes yours.

Try it yourself

import pandas as pd
import matplotlib.pyplot as plt
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans
from sklearn.metrics import silhouette_score
from scipy.cluster.hierarchy import linkage, fcluster

# One row per customer of the plant shop
plants = pd.read_csv("plants.csv")
features = ["orders_per_year", "avg_basket_eur", "sale_share_pct", "days_since_order"]

# The height where the owner cuts the shop's tree, read off its dendrogram (3 or more segments)
cut = float(input())

fig, ax = plt.subplots()
ax.set_title("Plant Shop Segments")
ax.set_xlabel("Orders per year")
ax.set_ylabel("Average basket (EUR)")

# Write your code below