dataquality 0.8.43

dataquality is a library for tracking and analyzing your machine learning models.

dataquality

The Official Python Client for Galileo.

Galileo is a tool for understanding and improving the quality of your NLP (and soon CV!) data.

Galileo gives you access to all of the information you need, at a UI and API level, to continuously build better and more robust datasets and models.

dataquality is your entrypoint to Galileo. It helps you start and complete the loop of data quality improvements.

ToC

• Getting Started
• Custom Integrations
• No labels? No problem
• Programmatic Access
• Contributing

Getting Started

Install the package.

pip install dataquality

Create an account at Galileo

Grab your token

Get your dataset and analyze it with dq.auto (You will be prompted for your token here)

import dataquality as dq

dq.auto(
    train_data="/path/to/train.csv",
    val_data="/path/to/val.csv",
    test_data="/path/to/test.csv",
    project_name="my_first_project",
    run_name="my_first_run",
)

☕️ Wait for Galileo to train your model and analyze the results.
✨ A link to your run will be provided automatically

Pro tip: Set your token programmatically for automated workflows

By setting the token, you'll never be prompted to log in

import dataquality as dq

dq.config.token = 'MY-TOKEN'

For long-lived flows like CI/CD, see our docs on environment variables

What kinds of datasets can I analyze?

Currently, you can analyze Text Classification and NER

If you want support for other kinds, reach out!

Can I use auto with other data forms?

auto params train_data, val_data, and test_data can also take as input pandas dataframes and huggingface dataframes!

What if all my data is in huggingface?

Use the hf_data param to point to a dataset in huggingface

import dataquality as dq

dq.auto(hf_data="rungalileo/emotion")

Anything else? Can I learn more?

Run help(dq.auto) for more information on usage
Check out our docs for the inspiration behind this methodology.

Can I analyze data using a custom model?

Yes! Check out our full documentation and example notebooks on how to integrate your own model with Galileo

What if I don't have labels to train with? Can you help with labeling?

We have an app for that! Currently text classification only, but reach out if you want a new modality!

This is currently in development, and not an official part of the Galileo product, but rather an open source tool for the community.

We've built a bulk-labeling tool (and hosted it on streamlit) to help you generate labels quickly using semantic embeddings and text search.

For more info on how it works and how to use it, check out the open source repo.

Is there a Python API for programmatically interacting with the console?

Yes! See our docs on dq.metrics to access things like overall metrics, your analyzed dataframe, and even your embeddings.

Contributing

Read our contributing doc!