What is Neptune?

Neptune is a collaboration hub for data science / machine learning teams. that focuses on three areas:

  • Track: all metrics and outputs in your data science or machine learning project. It can be model training curves, visualizations, input data, calculated features and so on.

  • Organize: automatically transform tracked data into a knowledge repository.

  • Collaborate: share, compare and discuss your work across data science project.

image

Neptune is lightweight

Neptune is built with the single design principle in mind: being lightweight. What does it mean in practice?

  • easy user onboarding: if you know how to use print() you will learn how to use it in no time.

  • 20-minute deployment: use SaaS, deploy on any cloud or own hardware (contact us to learn more).

  • Neptune fits in any workflow, ranging from data exploration & analysis, decision science to machine learning and deep learning.

  • Neptune works with common technologies in data science domain: Python 2 and 3, Jupyter Notebooks, R.

  • Neptune integrates with other tools like MLflow and TensorBoard.

Neptune’s focus: track, organize and collaborate

Track

Track all objects in the data science or machine learning project. It can be model training curves, visualizations, input data, calculated features and so on. Snippet below, presents example integration with Python code.

import neptune

neptune.init('shared/onboarding',
             api_token='eyJhcGlfYWRkcmVzcyI6Imh0dHBzOi8vdWkubmVwdHVuZS5tbCIsImFwaV9rZXkiOiJiNzA2YmM4Zi03NmY5LTRjMmUtOTM5ZC00YmEwMzZmOTMyZTQifQ==')
with neptune.create_experiment():
    neptune.append_tag('minimal-example')
    n = 117
    for i in range(1, n):
        neptune.send_metric('iteration', i)
        neptune.send_metric('loss', 1/i**0.5)
    neptune.set_property('n_iterations', n)

api_token belongs to the public user Neptuner. So, when started you can see your experiment at the top of experiments view.

Organize

Organize structure of your project:

  • code,

  • notebooks,

  • experiment results,

  • model weights,

  • meeting notes,

  • reports.

Everything is in one place, accessible from the app or programmatically. Neptune exposes Query API, that allows users to access their Neptune data right from the Python code.

Collaborate

Collaborate in the team:

  • share your experiments,

  • compare results,

  • comment and communicate your work,

  • use widgets and mentions to show your progress.

Speak Your language in our data-science focused interactive wiki!

image

Documentation contents