rel-trial: study outcome (Classification)

This example now materializes the official rel-trial tables from the relbench Python library and runs one GraphReduce instance for each official yearly training cut date, plus the validation and test cut dates.

from relbench_trial_study_outcome import run_rel_trial_study_outcome

df_train, df_val, df_test, in_time_auc, holdout_auc, n_features, materialized, target = run_rel_trial_study_outcome()
print(materialized)
print(target, in_time_auc, holdout_auc, n_features, len(df_train), len(df_val), len(df_test))

Current implementation:

• examples/relbench_trial_study_outcome.py

Latest Tested Result

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

• in-time AUC: 0.6208
• holdout AUC: 0.6434
• feature count: 282
• temporal instances: 21 (19 train, 1 validation, 1 test)
• rows: train 11,994, validation 960, test 825
• runtime: approximately 45s on the local development machine

Interactive Runner

Save API URL Run rel-trial study outcome

Idle