Skip to content

AGENTS.md

Latest commit

 

History

History
57 lines (47 loc) · 3.47 KB

AGENTS.md

File metadata and controls

57 lines (47 loc) · 3.47 KB
Raw

Agent Guide

Scope: this file applies to the whole repository.

Repository Map

This is an R/Quarto research project for the paper "RBA vs Machine", comparing Reserve Bank of Australia forecasts with machine-learning and benchmark forecasts.

  • run_all.R is the master reproducibility script. It sources the numbered R pipeline in order.
  • R/00_setup.R installs/loads required R packages, creates output directories, and sources shared helpers.
  • R/01_data_pull.R pulls ABS, RBA, FRED, and Yahoo Finance data using series lists in data/config/.
  • R/02_stationarity.R applies stationarity transformations and writes processed panels/tests.
  • R/03_rba_forecasts.R builds the RBA historical forecast and forecast-error data.
  • R/04_ml_models.R and R/04_ml_models_fast.R train the forecast models. The master pipeline uses the fast version.
  • R/05_evaluation.R evaluates forecast accuracy and robustness.
  • R/06_exhibits.R writes publication figures to figures/ and tables to tables/.
  • R/functions/ contains shared helpers for data pulls, transformations, ML fitting, and evaluation.
  • documents/ contains the Quarto/R Markdown manuscripts, bibliography, CSL file, LaTeX header, and rendered PDFs.
  • data/config/ contains committed source-series definitions. Treat these CSVs as inputs.
  • data/processed/ contains committed processed summaries. data/output/ and data/raw/ are generated/ignored.
  • figures/ and tables/ contain rendered exhibits used by the manuscript.
  • Python/ contains exploratory notebook/utilities for importing data and stationarity checks; the canonical pipeline is currently in R.
  • resources/ contains research notes and reference PDFs.

Common Commands

  • Run the full pipeline: 
    Rscript run_all.R
  • Render the main manuscript: 
    quarto render documents/RBAvsMachine.qmd
  • Re-run only evaluation and exhibits after model outputs exist: 
    Rscript R/run_eval_and_exhibits.R
  • Check R syntax without running the full pipeline: 
    Rscript -e "invisible(lapply(list.files('R', pattern = '\\\\.R$', recursive = TRUE, full.names = TRUE), parse))"

Data And Outputs

  • The data pull hits external services and may need network access plus a valid FRED_API_KEY. R/01_data_pull.R falls back to a hard-coded key if the environment variable is missing; prefer setting FRED_API_KEY in the environment for new runs.
  • Do not commit generated data/output/, data/raw/*.csv, .RData, .Rhistory, .Rproj.user/, LaTeX intermediates, or training logs.
  • Preserve committed configuration files in data/config/; changes there alter the empirical information set.
  • figures/ and tables/ are committed outputs. If code changes affect exhibits, regenerate and review them before committing.

Editing Guidelines

  • Keep the numbered R pipeline modular. Prefer editing the relevant step and shared helper rather than adding hidden side effects elsewhere.
  • Use here::here()/here::i_am() patterns already present in the R code instead of hard-coded absolute paths.
  • Be careful with long-running or networked steps. For small edits, prefer syntax checks or targeted scripts before running the full pipeline.
  • Treat the Quarto manuscripts as source files; generated .tex and *_files/ artifacts are ignored unless explicitly requested.
  • documents/DatabasePaper.qmd is draft material and currently contains merge-marker text in the committed file. Do not clean or rewrite it unless the task asks for that document.