First zoo run

You are about to run a public bake-off: several readout discriminators, one frozen recipe, one report. By the end of this sitting you will have plots under output/ and a ranking you can open in a lab meeting.

Complete Quickstart first if the package is not yet installed or if you have not verified arcade list.

Mission. Install the zoo extras, link the Readout 2019 files the bundled configs expect, launch configs/paper_all_methods_benchmark.yaml, then open the summary report and name a provisional winner under the shared recipe.

        flowchart LR
  A[Install extras] --> B[Link paper data]
  B --> C[arcade list]
  C --> D[arcade run zoo YAML]
  D --> E[Open summary report]
    

• Preparevenv, extras, data symlinks
• Launchone YAML, many methods
• Inspectreport table + plots
• Extendyour traces or a custom class

Install the extras used by the public zoo

cd ARCADE
pip install -e ".[torch,signature]"
bash scripts/link_paper_data.sh
arcade list

The link_paper_data.sh script symlinks Readout 2019 HDF5 files into data/ so the bundled paper configs resolve. Verify the environment with:

python -c "import arcade; print('arcade import ok')"
arcade describe threshold

You should see a non-empty classifier list from arcade list and a short description for threshold. Treat that as your green light before the long run.

Optional extras you may need later

Extra	Command	Enables
Core only	pip install -e .	Threshold and scikit-learn baselines without PyTorch
torch	pip install -e ".[torch]"	FNN, HERQULES, leakage, multilevel, transformer
signature	pip install -e ".[signature]"	Path-signature classifier via iisignature
hardware	pip install -e ".[hardware]"	HLS4ML-oriented helpers
qec	pip install -e ".[qec]"	Stim-based logical error and FTQC curves
dev	pip install -e ".[dev]"	pytest and ruff
docs	pip install -e ".[docs]"	Sphinx build via make -C doc html
export	pip install -e ".[export]"	PDF figure export with PyMuPDF
All	pip install -e ".[all]"	Full dependency set

Run the public zoo

arcade run configs/paper_all_methods_benchmark.yaml

While it runs, the pipeline walks data preparation, feature build, classify for every name in classifiers.run, optional hardware estimation, and report writing. Outputs land under output/. Expect per-method plots and a summary report when visualization is enabled in that configuration.

Same run from Python

YAML is the supported public path. ADC file paths and demodulation settings live in the config:

import arcade

arcade.run("configs/paper_all_methods_benchmark.yaml")

Useful companion commands while you wait or after the run finishes:

arcade list
arcade describe fnn

Point ARCADE at your own traces

Once the public zoo works, swap the data path and shrink the method list for a faster personal bake-off:

data:
  path: /path/to/your_traces.h5
  format: hdf5
  num_qubits: 5

classifiers:
  run: [threshold, lda, qda, svm, fnn, herqules]

sweep:
  enabled: true

visualization:
  output_dir: output/my_run

arcade run path/to/my_config.yaml

Fill demodulation and split fields so they match your file contract. Keep those fields identical for every method in the run. See Compare classifiers for a fuller bake-off recipe, and Build your own classifier when you want a custom method beside the zoo.

Next

Continue with Read the report to interpret Accuracy, F5Q, LUT, and latency columns. That tutorial is where the ranking becomes a decision rather than a pile of files.