CLI reference¶

Every subcommand of srn forwards to a Python module under swissrivernetwork.benchmark. Forwarding preserves __main__ semantics so the drivers can still be invoked via python -m if you prefer.

Command	Driver
`srn prepare-data`	`swissrivernetwork.benchmark.data_preparation`
`srn tune -m <method> -g <graph> -n <n> -wl <wl> [-pe <pe>]`	`swissrivernetwork.benchmark.ray_tune`
`srn evaluate`	`swissrivernetwork.benchmark.ray_evaluation`
`srn sweep`	`swissrivernetwork.benchmark.run_win_len_sweep`
`srn train-single`	`swissrivernetwork.benchmark.train_single_model`
`srn train-isolated`	`swissrivernetwork.benchmark.train_isolated_station`
`srn app gradio`	`swissrivernetwork.app.gradio_app`
`srn app streamlit`	Streamlit launches `swissrivernetwork/app/streamlit_app.py`
`srn version`	Print the installed package version

Driver-specific flags go after --:

uv run srn tune -- -m transformer_embedding -g swiss-2010 -n 200 -wl 90 -pe rope

Common flags¶

-m / --method — one of lstm, lstm_embedding, graphlet, stgnn, transformer, transformer_graphlet, transformer_embedding, transformer_stgnn.
-g / --graph — one of swiss-1990, swiss-2010, zurich.
-n / --num-trials — Ray Tune trial count.
-wl / --window-len — temporal window length (days).
-pe / --positional-encoding — sinusoidal, learnable, rope.

See the .claude/skills/run-benchmark/SKILL.md playbook for the full reference including the environment variables that must stay set (notably RAY_CHDIR_TO_TRIAL_DIR=0).