rel-avito: user clicks (Classification)

This runner now uses the official relbench dataset loader and materializes the Avito tables locally before GraphReduce runs. The example no longer depends on the legacy S3 parquet mirror.

from relbench_avito_user_clicks import run_rel_avito_user_clicks

df, auc, n_features, materialized, target = run_rel_avito_user_clicks()
print(materialized)
print(target, auc, n_features, len(df))

Current implementation:

• examples/relbench_avito_user_clicks.py
• examples/relbench_avito_common.py

Latest Tested Result

Latest local run from tests/data/relbench/run_reports/relbench_results.json:

• test ROC AUC: 0.7895
• test average precision: 0.2750
• feature count: 281
• rows: train 59454, validation 21183, test 47996
• runtime: 30.0s (2026-04-09T22:24:39Z)

Interactive Runner

Save API URL Run rel-avito user clicks

Idle