Verification

Before you launch a multi-method zoo run that can take hours, confirm that the environment resolves imports, that the console script is on your PATH, and that at least one simple classifier can be constructed from the registry. These checks catch missing extras and broken editable installs early.

Import and console script

python -c "import arcade; print(arcade.__version__ if hasattr(arcade, '__version__') else 'ok')"
arcade list

arcade list should print a table or list of registered classifier names. If the command is not found, reinstall with pip install -e . from the repository root inside your activated virtual environment.

Describe a baseline method

arcade describe threshold

You should see a short description of the threshold classifier. If the name is unknown, the registry did not load that module. Check that you are in the correct repository and that the editable install points at this tree.

Smoke-run a tiny configuration

Prefer a configuration that trains only one fast method and a small number of epochs when you are validating an install. For example:

arcade run configs/examples/threshold_only.yaml

If your tree uses a different example path, pick any YAML that lists a single classical classifier and a small dataset subset. The run should create an output directory, write a report or summary artifacts, and exit with status zero. Failures at this stage are usually missing data symlinks, missing HDF5 files, or optional dependencies that a config assumed were installed.

Link paper data when needed

Many paper configs expect files under data/. From the repository root:

bash scripts/link_paper_data.sh

If the script cannot find the source dataset location on your machine, read its comments and the dataset notes in Guides before retrying a paper config.

Optional dependency checks

Capability	Quick check
PyTorch	`python -c "import torch; print(torch.__version__)"`
Path signatures	`python -c "import iisignature"`
QEC stack	`python -c "import stim; import pymatching"`
Docs build	`make -C doc html`

Install the matching extra from Installation if any of these imports fail and you need that capability.

What success looks like

A healthy Quickstart ends with arcade list working, a describe call for a known method succeeding, and a short config run writing outputs without traceback. At that point you are ready for Guides if you have a goal, for API Reference if you need exact names, or for Tutorials if you want a paced first zoo run.