---

name: recommendation-and-analysis description: "Turn fitted Surprise models or prediction lists into ranked recommendations, precision/recall@k summaries, and safe dump/load roundtrips." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

Recommendation and Analysis

Use this sub-skill when you already have a fitted Surprise model, a Trainset, or a list of Prediction objects and you need to:

• build an anti-testset from known users and items,
• rank candidate items into top-N recommendations,
• compute precision@k / recall@k over prediction lists,
• serialize predictions or fitted algorithms with dump.dump / dump.load,
• interpret Prediction records and raw ids safely.

What this route covers

• Trainset.build_anti_testset(fill=None) and fill behavior
• stable top-N ranking over Prediction outputs
• precision/recall@k conventions, including zero-division handling
• dump.dump and dump.load roundtrips using temp files
• raw-id handling for recommendation output

What this route excludes

• model fitting or tuning
• cross-validation and search
• algorithm internals beyond cross-links and the minimal Prediction / AlgoBase contract

Entry points

• references/recommendation.md for usage patterns and output contracts
• references/troubleshooting.md for common failure modes
• scripts/top_n_smoke.py for anti-testset ranking and tie ordering
• scripts/precision_recall_smoke.py for zero-division cases in precision/recall@k
• scripts/serialize_smoke.py for temp-file dump/load roundtrip

Use when

• you already have algo.test(...) output and want ranked recommendations
• you need per-user precision/recall@k from a Prediction list
• you need to persist predictions or a fitted algorithm without keeping the original process alive

Handoff notes

If you still need trainset creation or model fitting, hand off to the data-loading or prediction-algorithms route first, then return here with a fitted model or predictions.