Skip to content

Latest commit

 

History

History
53 lines (41 loc) · 4.25 KB

README.md

File metadata and controls

53 lines (41 loc) · 4.25 KB

Research Scripts

These command-line scripts support the broader, historical research workspace. They are not required to read or validate the submitted dissertation. For the preserved submission workflow, start with make validate.

Public and local utilities

Task Entry point Notes
Build the public benchmark dataset build_labeled_dataset.py Rebuilds the tracked PhraseBank/FiQA derivative; do not edit the CSV by hand.
Check repository boundaries check_project.py Verifies the submission inventory, public-file policy, and documentation links.
Rebuild the GitHub share card build_social_preview.py Uses the locked plotting extra; presentation artwork only, with no data or provider calls.
Inspect UCL GPU availability ucl_gpu_status.py Read-only SSH status probe; requires UCL access.
Run the benchmark TUI on Windows run_tui.ps1 Convenience launcher for the root Python package.

Historical experiment tooling

Collection and preparation

Workflow Entry points Guide
LSEG collection cleanup and merging build_lseg_clean_csvs.py, merge_lseg_headline_collections.py LSEG pipeline
Materiality sample and story bodies prepare_lseg_materiality_pilot.py, fetch_lseg_materiality_bodies.py Research archive
Strategy price panel build_strategy_price_panel.py Strategy-research pipeline
FNSPID source retrieval and audit download_fnspid.py, audit_fnspid_news.py Data and rights

Scoring and remote execution

Workflow Entry points Guide
Historical LSEG baseline study run_lseg_headline_study.py Baseline design
OpenRouter validation and LSEG scoring run_openrouter_gemma4_validation.py, run_openrouter_lseg_scoring.py Validation and authorisation record
UCL local-model execution ucl_run_lseg_baselines.sh, ucl_label_headlines.sh, ucl_ollama_server.sh UCL GPU labelling
Backward-replication status and launch audit_lseg_backward_pipeline_status.py, watch_and_launch_lseg_finbert_ucl.py Backward-replication runbook

Reports and derived views

Output Entry point Guide
Tail-risk factorial comparison build_fnsipid_tail_risk_factorial.py Variant comparison
Frozen strategy dashboard build_recent_news_dashboard.py Mid-cap result
Week 6 trade explorer build_week6_trade_dashboard.py Trade-explorer guide
Week 7 receipt-aligned P/L build_week7_receipt_pnl.py Receipt-aligned guide

These are historical entry points, not a suggested sequence to rerun. Scripts that fetch data, score text, or launch remote jobs may use external services, credentials, licensed inputs, substantial compute, or a recorded authorisation flag. Read the linked protocol and the script's --help output before running one.

Generated outputs belong in the ignored data, result, report, or experiment-output locations already defined by the project. Preserve existing manifests and run folders; the research tools generally create a new artifact or refuse to overwrite one.