Skip to main content

Find your way in a maze of experiments

Project description

Ariadne is a lightweight and minimal Python library for managing computational experiments and their results. To install, run:

uv add ariadne-lib

or

pip install ariadne-lib

[!WARNING] This library is developed for personal use. Anyone is welcome to use it and give feedback, but be ready for sharp edges and breaking changes.

Basic Usage

The primary goal of Ariadne is to help keep track of what you did and what the results were. The rapid prototyping workflow that research encourages often leads to issues with reproducibility. Plots and results can become stale and lack provenance, results may often become overwritten and lost.

The solution Ariadne provides is the simplest one possible: assign each experiment to a different unique folder, and provide simple utilities for creating and managing these folders, as well as saving useful metadata (like timestamps, notes, and configs). The primary experiment management happens in the Theseus class. Calling start or resume returns a id for that experiment, along with the dedicated folder that everything should be stored under.

from ariadne import Theseus

experiment = Theseus(db_path="experiments.db", base_dir="results")
experiment_id, experiment_folder = experiment.start(name="resnet", run_config={"architecture": "resnet-50"}, notes="try resnet instead of mlp")

for epoch in range(len(50)):
    x = [0, 1]
    yhat = [0, 0]
    y = [1, 1]
    loss = 1
    experiment.log({"epoch": epoch, "loss": loss})

    plt.plot(x, yhat, "-")
    plt.plot(x, y, "--")
    plt.savefig(experiment_folder / f"predictions{epoch}.png")

Ariadne then provides utilities for querying the experiment database by name (or an exact match by id), to quickly associate an experiment with its dedicated folder on disk.

experiment_id, experiment_folder = experiment.get("resnet") # searches by substring match
experiment_id, experiment_folder = experiment.peek() # last run experiment
experiment_id = 3
experiment_folder = experiment.get_by_id(experiment_id) # or exact id match

img = plt.imread(experiment_folder / "predictions49.png")
plt.imshow(img)

Finally, Ariadne provides a simple cli for querying and summarizing experiments. Run ariadne --help for more information.

FAQ:

Q: Why the name?

A: My project directories often end up evolving into a tangled mess of results, plots, and files, not unlike a maze. In greek mythology, Ariadne's red string was the crucial key for helping Theseus escape the labyrinth of Crete.

Q: Why not weights and biases, tensorboard, or other experiment management tools?

A: You probably should use them instead if you can! But here's a brief reason why I wrote this library instead:

  • Experiments are stored locally and a simple CLI is provided. No need for a web server or a browser.
  • Because all Ariadne does is handle creation and management of experiment folders, it composes well with other libraries. Your logging, plot creation and saving, and checkpointing code can remain exactly the same - just prepend the experiment folder in front of the save path.
  • I simply don't need most of the features these other experiment management frameworks offer.

Q: Why not git lfs?

A: Git LFS is great for versioning large files. But, when you are rapidly iterating and creating throwaway experiments that you might not need to share, constantly writing and overwriting rich data like plots makes for horribly ugly diffs, and can bloat your cloud storage.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ariadne_lib-0.1.6.tar.gz (144.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ariadne_lib-0.1.6-py3-none-any.whl (9.3 kB view details)

Uploaded Python 3

File details

Details for the file ariadne_lib-0.1.6.tar.gz.

File metadata

  • Download URL: ariadne_lib-0.1.6.tar.gz
  • Upload date:
  • Size: 144.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.4.24

File hashes

Hashes for ariadne_lib-0.1.6.tar.gz
Algorithm Hash digest
SHA256 b4a918196eea62eb2d311415893ce827424925e6c444c8aa3e5a453fd4ff1de8
MD5 3f0fd0446a513ca86bc9bcbebc75fac9
BLAKE2b-256 865d723feccc1825e429a1888968bf66d5233b0aa774a70fb003ba4efd4cce8f

See more details on using hashes here.

File details

Details for the file ariadne_lib-0.1.6-py3-none-any.whl.

File metadata

File hashes

Hashes for ariadne_lib-0.1.6-py3-none-any.whl
Algorithm Hash digest
SHA256 6cdf3dfe8d305d6f40646f60dbb4d3ecb927c8db855bfa7472f37ed3fd3d2f92
MD5 151449e3f5d777d3a57044123fdf5ece
BLAKE2b-256 cec6abb8f880314523a2c2fe3142c3216f4884bb6d87590f5cee203f59bbdc24

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page