BayesMMfwd.jl

Forward model for mm-source catalogs, with a Julia baseline path and a Reactant tracing/compiled path.

The main entrypoint is:

• gen_pysides_from_original.jl

It loads a SIDES catalog, generates SFR/magnification/flux columns, and runs benchmark timings.

Repository Layout

• gen_pysides_from_original.jl: main driver and benchmark harness.
• gen_sfr_props.jl: SFR-related computations (gen_sfr_props2 is currently used by the driver).
• gen_magnification.jl: magnification generation.
• gen_fluxes.jl: SED/flux/LFIR generation.
• load_params.jl: parser for SIDES_from_original.par.
• load_sides_csv.jl: CSV loader for SIDES catalog inputs.
• data/: input catalogs (SIDES_Bethermin2017_short2.csv, SIDES_Bethermin2017_short.csv).
• SIDES_from_original.par: parameter file consumed by the pipeline.
• SED_finegrid_dict.h5, LFIR_LIR_ratio.h5, Psupmu_table_Bethermin17.txt: required data assets.

Prerequisites

• Julia with project environment support.
• Files above available at their expected paths.
• For Reactant local development in this repo, Reactant.jl/ is a symlink to a local checkout.

Quick Start

From repo root:

julia --project=. gen_pysides_from_original.jl

This command:

1. Loads parameters and catalog input.
2. Preallocates output columns on the catalog.
3. Runs benchmark process paths.
4. Writes a formatted benchmark summary to a text file.

Choosing Dataset Size

In gen_pysides_from_original.jl, update csv_idl_path near the bottom:

• Small/debug run: data/SIDES_Bethermin2017_short2.csv
• Large benchmark run: data/SIDES_Bethermin2017_short.csv

Benchmark Outputs

The script writes a formatted benchmark report via format_benchmark_report(...).

Typical output files used in this repo:

• benchmark_results_small.txt
• benchmark_results_large.txt

The report includes:

• dataset path
• backend (cpu)
• seed
• timing table
• derived ratios

Configuration and Reproducibility

Important knobs in gen_pysides_from_original.jl:

• BENCHMARK_SEED = 1234
• Reactant.set_default_backend("cpu")
• parameter file path: "SIDES_from_original.par"

If you are comparing runs, keep seed and dataset fixed.

Troubleshooting (Reactant)

Scalar indexing is disallowed

Use @allowscalar only where scalar get/set is required (for example explicit index reads/writes inside traced loops).

findall on traced boolean masks fails

Prefer mask-based formulations with ifelse.(mask, a, b) rather than findall(mask) + indexed writes.

Trace/compile timings look too small

Compiled/traced calls can dispatch asynchronously; if timings look unrealistically tiny, ensure your timing method includes synchronization.

Missing data file/path errors

Check:

• SIDES_from_original.par
• data/SIDES_Bethermin2017_short*.csv
• SED_finegrid_dict.h5
• LFIR_LIR_ratio.h5
• Psupmu_table_Bethermin17.txt

Development Notes

• This repo currently contains several MWE scripts (mwe_*.jl) used for Reactant debugging.
• Keep helper scripts focused; most orchestration should stay in gen_pysides_from_original.jl.