rkw/bank-fraud-baf-lakehouse
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
20 Commits 1 Branch 0 Tags
5091aec53cead97f7a285edfb087c8e9e69f1661
Commit Graph

4 Commits

Author SHA1 Message Date
Rob Wiederstein
df978d042f Refactor bucket structure: baf-fraud/ prefix under lake bucket 
All functions now default to bucket_name = "lake" with "baf-fraud/"
prepended to all layer prefixes, matching the contemporary lakehouse
naming convention (one bucket per environment, project as prefix).

Migration: copy baf-fraud/ data to lake/baf-fraud/ on analyticsvm,
update BAF_BUCKET env var from "baf-fraud" to "lake".

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-02-22 05:36:25 -05:00
Rob Wiederstein
b38892f49e Refactor: consistent naming across functions, targets, and pkgdown 
Functions: prepare_eda_recipe -> build_eda_recipe,
           create_efficiency_plot -> plot_efficiency,
           format_class_imbalance_tourney_gt -> format_tournament_gt

Targets: model_inputs_prefix -> baf_model_input_prefix,
         tbl_fraud_by_month_data -> fraud_by_month_summary,
         model_diag -> diag_fit, winning_params -> best_params,
         production_recipe_blueprint -> prod_recipe,
         final_eval_data -> test_predictions

pkgdown: restructured reference index into 6 logical sections,
         removed stale names and development comments.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-02-22 03:52:34 -05:00
Rob Wiederstein
f47b2e1be2 Add tune_lgbm() and wire hyperparameter tuning into DAG 
Converts scratch/tune_model.R into a pure tune_lgbm() function,
replacing hardcoded winning_params with a fully automated tar_target.
Best params (trees=844, depth=3, lr=0.0204, min_n=389) now flow
reproducibly into evaluate_final_model() and train_production_model().
PR-AUC improved from 0.165 to 0.198.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-02-22 03:25:35 -05:00
Rob Wiederstein
33d0fc31c7 Initial commit: BAF Lakehouse fraud detection pipeline 
End-to-end LightGBM fraud detection pipeline built as an R package,
orchestrated by targets with data stored in MinIO via Apache Arrow.
Includes 6-layer Lakehouse architecture, class imbalance tournament,
formally tuned hyperparameters (PR-AUC 0.198), and Quarto RevealJS slides.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-02-21 21:19:09 -05:00