Deborah.DeborahThreads

module DeborahThreads

Deborah.DeborahThreads — Threaded dispatcher for multi-job Deborah.DeborahCore runs.

Deborah.DeborahThreads parallelizes multiple Deborah.DeborahCore jobs (different LBP/TRP partitions, or multiple model/config variations) using Julia Threads. It parses a single TOML that enumerates label/train pairs, expands them into per-job arguments, writes per-job configs, and launches Deborah.DeborahCore.DeborahRunner.run_Deborah concurrently with simple batch scheduling.

Scope & Responsibilities

• Config expansion: Parse a multi-job TOML and materialize one job per (LBP, TRP).
• Per-job config writer: Generate TOML dictionaries/files for each job.
• Threaded execution: Dispatch jobs in batches of Base.Threads.nthreads(); wait per batch.
• Run logging: Record start/end timestamps and errors per job into run_*.log.

Key Components

• DeborahThreadsRunner.JobArgsDeborah: immutable container of a single Deborah.DeborahCore job's arguments.
• DeborahThreadsRunner.parse_config_DeborahThreads → Vector{JobArgsDeborah}: Expand the global config into concrete jobs.
• DeborahThreadsRunner.generate_toml_dict → Dict: Build a TOML-ready dictionary for one job (data / bootstrap / jackknife / abbreviation).
• DeborahThreadsRunner.run_one_job: Write per-job TOML, call Deborah.DeborahCore.DeborahRunner.run_Deborah, and log results.
• DeborahThreadsRunner.run_DeborahThreads: High-level entry point: parse → batch → spawn → wait.

File/Path Conventions

• Output base: <location>/<analysis_header>_<ensemble>/<analysis_header>_<ensemble>_<X_Y_or_code>_<model_suffix>/
• Per-job directory: .../<overall_name>/
• Per-job config: config_Deborah_<overall_name>.toml
• Per-job log: run_Deborah_<overall_name>.log

Minimal Usage

julia> using Deborah
julia> run_DeborahThreads("config_DeborahThreads.toml")

Notes

• Abbreviation support: feature/target names can be encoded via Deborah.Sarah.StringTranscoder.
• Batching equals Base.Threads.nthreads(); adjust JULIA_NUM_THREADS to change parallel width.
• Column indices in the input spec are $1$-based; they are propagated into per-job TOMLs.
• Exceptions are caught per job and written into the corresponding .log file.