An intuitive way to build models

PerceptiLabs is a dataflow driven, visual API for TensorFlow that enables developers to work more efficiently with machine learning models and to gain more insight into their models. It wraps low-level TensorFlow code to create visual components, which allows users to visualize the model architecture as the model is being built.

This visual approach lowers the barrier of entry for beginners while providing researchers and advanced users with code-level access to their models.

A visual API for TensorFlow

As a visual API, PerceptiLabs sits on top of TensorFlow and other APIs:

PerceptiLabs wraps low-level TensorFlow code to create visual components, so you’ll see your model’s architecture as you build.

Real-time insights

See real-time analytics and granular previews of output from each model component. You can easily track and understand the gradients’ behavior, perform real-time debugging, and see where to optimize your model.

Keep track of models and share on GitHub

PerceptiLabs lets you manage multiple models, compare them, and easily share the results back to your team. Export your model as a TensorFlow model or as a Jupyter Notebook.

Features

The following are some of the key features of PerceptiLabs:

• Shows the model architecture with output visualizations for each component
• Shows granular visualizations during the modeling phase, run-time, and testing
• Shows warnings and recommendations for debugging and model building
• Automatically suggests configs/settings and hyperparameters
• Provides access to the underlying code for editing in the tool
• Includes model templates for common machine learning problems
• Performs dimension and I/O shape fitting
• Includes a model registry to easily keep track of models and experiments
• Includes data and model version control to reproduce experiments and go back in time
• Can perform distributed training over all available GPUs
• Performs different tests to try out the model before pushing it to production

Quickstart

PerceptiLabs is offered as a free Python package (hosted on PyPI) for everyone to use.

Install it:

pip install perceptilabs

Run it:

perceptilabs

This will run the PerceptiLabs kernel locally on your machine and launch its user interface in your web browser.

Documentation →

How to cite us

If you're writing a paper or article about a project that used PerceptiLabs, we'd love it if you cited us. Here's a generated BibTeX citation for our website that will help point people to our tools:

@misc{pl,
title = {Visual Machine Learning Modeling with PerceptiLabs},
year = {2020},
note = {Software available from www.perceptilabs.com},
url={www.perceptilabs.com},
author = {Martin Isaksson and Robert Lundberg},
}

Community

Got questions, feedback, or want to join a community of machine learning practitioners working with exciting tools and projects? Check out our forum and Slack Channel!

Follow us on Twitter.