Level 2 · Find the slow movers in pandas

What you are checking, and why

Before ranking anything, decide which products can fairly be judged. Three groups behave differently:

• Retired products are already leaving the range; they belong to a different review.
• New in 2025 products had only part of the year on sale. A lantern launched in August cannot be compared with one that sold all summer.
• Established products were on sale for all twelve months: these are the ones to rank.

Step 1 starts from products and merges the 2025 sales onto it with how="left", so a product that sold nothing would still be there, with NaN units. np.select then labels each product, taking the first condition that is true:

p["product_group"] = np.select(
    [~p["is_active"], p["launched_on"] >= "2025-01-01"],
    ["retired", "new in 2025"],
    default="established",
)

In the aggregation, ("units", "count") counts the products that sold, and a small lambda counts missing unit costs, because a slow product with an unknown cost is one whose margin nobody can judge.

Ranking inside each category

Step 2 needs the category’s total on every product row. transform returns one value per row, aligned with the original table, where a plain sum() would return one row per category:

t["share"] = t["revenue"] * 100 / t.groupby("category")["revenue"].transform("sum")
low = t.sort_values(["category", "revenue"]).groupby("category").head(3)

head(3) after a sort is the pandas way of saying “the first three per group”. Step 3 uses the same trick, then sums the weakest three and divides by the category total.

Reading the result

category	weakest product	units_2025	category_share_pct
Furniture	Upland Oak Bookcase	9	0.7
Textiles	Spruce Wool Duvet Cover	25	0.7
Outdoor	Orchard Aluminum Adirondack Chair	93	1.0
Lighting	Fern Rattan Desk Lamp	13	1.2

Revenue share and units together tell the story. The bookcase and the desk lamp sold almost nothing; the Adirondack chair sold 93 units but at $51.99 each, so it barely registers in a category of patio sets and grills.

Pitfalls

• is_active is a boolean column, so ~p["is_active"] flips it; p["is_active"] == "false" matches nothing.
• groupby().head(3) keeps the first three rows in the current order, so sort first.
• A share of revenue says nothing about profit. Seven established products have no unit cost, so their margin is unknown.