Feature Engineering: Discretization & Binning (pd.cut vs pd.qcut)

From Numbers to Tiers: Turning Continuous Spend into Business Buckets Your stakeholder never says "Show me customers who spent $437.20 vs $441.80" — they say "How many VIPs do we have?" or "Who is in the top 20%?" . To answer, you must turn a continuous number like total_amount into a handful of labeled tiers . Pandas gives you two knives for that — cut and qcut — and they slice the same data in completely different ways. This article shows both on the same 10 customers so the difference is obvious.

1. The Data: 10 Customers Sorted by Total Spend Same customers as the RFM article — one row per customer, sorted low → high. This ordering makes both binning methods visible:

customer_id total_amount ($) Same value, what you will do with it C007 18 Small spenders C003 90 C009 210 Middle C004 680 C005 890 Big — but gaps vary wildly (890→2740 in 3 steps) C001 1,274 C002 1,380 C010 2,740 C008 8,500 Single huge outlier — 3× the next C006 12,000 Extreme outlier — would dominate any fixed-width bin The spread is 18 → 12,000 with one monster at 12K. Watch how each method handles it.

2. Method A — pd.cut : Same Dollar Width Per Bucket (Business Rules) cut chops the value range into equal-width intervals. You pick the dollar boundaries — like pricing tiers on a website.

bins = [0, 500, 1500, 5000, float('inf')] # dollar cutoffs you choose labels = ['Low', 'Medium', 'High', 'VIP'] # one label per interval

df['spend_tier'] = pd.cut(df['total_amount'], bins=bins, labels=labels) print(df[['customer_id','total_amount','spend_tier']]) # customer_id total_amount spend_tier # 0 C007 18 Low # 1 C003 90 Low # 2 C009 210 Low # 3 C004 680 Medium # 4 C005 890 Medium # 5 C001 1274 Medium # 6 C002 1380 Medium # 7 C010 2740 High # 8 C008 8500 VIP # 9 C006 12000 VIP # Counts: Low=3, Medium=4, High=1, VIP=2 → buckets have VERY different sizes! Bucket Dollar range Who lands here Count Low 0 – 500 18, 90, 210 3 Medium 500 – 1,500 680, 890, 1,274, 1,380 4 High 1,500 – 5,000 <td style