rel-hm: user churn (Classification)
This example now pulls rel-hm from the relbench Python library and materializes the official tables locally before GraphReduce runs. The old S3 parquet download path has been removed.
from relbench_hm_user_churn import run_rel_hm_user_churn
df, auc, n_features, materialized = run_rel_hm_user_churn()
print(materialized)
print(auc, n_features, len(df))
Current implementation:
Current RelBench-aligned dates:
- cut date:
2020-09-14 - lookback start:
2019-09-07
Latest Tested Result
Latest local run from tests/data/relbench/run_reports/relbench_results.json:
- test ROC AUC:
0.6952 - test average precision:
0.9045 - feature count:
218 - rows: train
80948, validation76556, test74575 - runtime:
36.4s(2026-04-09T22:51:48Z)
Interactive Runner
Idle