Skip to content

cli_reference.md

Latest commit

 

History

History
95 lines (71 loc) · 2.81 KB

cli_reference.md

File metadata and controls

95 lines (71 loc) · 2.81 KB
Raw

Command-line reference

Install the package in editable mode before using the module or console entry point:

python -m pip install -e ".[dev]"

These two forms are equivalent:

python -m bnc_repro.cli --help
bnc-repro --help

train

Run every grid item in one YAML configuration.

python -m bnc_repro.cli train --config CONFIG [--output OUTPUT_ROOT]
Argument Required Meaning
--config Yes YAML profile loaded and validated by bnc_repro.config
--output No Root for generated run directories; config/default is used when omitted

The loader validates selected experiment-specific values for formal profiles, but it does not bind every protocol name or architecture list. The current unified CLI does not expose grid overrides or resume. Review the YAML and use a separate formal: false pilot config rather than changing a manuscript profile in place.

aggregate

Combine run-level metrics.csv files below a selected root.

python -m bnc_repro.cli aggregate \
  --figure FIGURE \
  --runs-root RUNS_ROOT \
  --output OUTPUT

FIGURE is one of fig1, fig2, fig_s1, fig_s2, fig_s3, or fig_s4. The command recursively reads every metrics.csv below --runs-root and does not inspect status.json; point it only at compatible runs whose status is complete. Figure 1 has no generic aggregate path because its PCA requires a trained checkpoint. Figure 2 writes a directory of raw, aggregate, and onset tables; the other supported aggregators write the requested CSV path.

plot

Create PNG, PDF, and SVG from one compatible data table.

python -m bnc_repro.cli plot \
  --figure FIGURE \
  --data INPUT_TABLE \
  --output OUTPUT_STEM \
  [--font-family Arial]

--output is a stem, not a directory. For example, --output figures/custom/fig_s3 creates fig_s3.png, fig_s3.pdf, and fig_s3.svg.

validate

Validate supplied paper data for one target or all targets.

python -m bnc_repro.cli validate \
  --figure {fig1,fig2,fig_s1,fig_s2,fig_s3,fig_s4,all} \
  [--data paper_data]

The command prints JSON. A deliberate blocked state is not the same as an exception: Figure 1 and the full Figure 2 plot have documented missing inputs.

Wrapper scripts

Script Purpose
scripts/run_smoke.py Train four tiny CPU runs, aggregate, and plot
scripts/validate_reference_data.py Validate all supplied paper data and write a report
scripts/plot_all_paper_data.py Plot every available supplied-data target and write statuses
scripts/reproduce_fig*.py Preselect one figure for the plot subcommand
scripts/check_tutorials.py Verify repository-local Markdown links

Run commands from the repository root so relative config and data paths match the documented examples.