TensorSpace.js provides an open source browser-based neural network data visualization framework to complement the growing machine learning landscape by supporting pre-trained models created with TensorFlow.js, Keras, or TensorFlow.
The project helps front-end developers visualize machine learning model structures as well as the processes of internal feature abstractions, intermediate data manipulations, and final inference generations.
TensorSpace.js leverages Three.js for its underlying 3D drawing API. The project adds data visualization of functional and sequential models including LeNet, AlexNet, YOLOv2, ResNet-50, Vgg16, ACGAN, MobileNetv1, Inceptionv3, and more. Examples of each model may be viewed at the TensorSpace.js playground.
A TensorSpace layer provides a container to show 3D visualizations for the internal layer data and structure, including features such as dense, flatten, reshape, pooling and more in a manner intended to be familiar to those working with machine learning APIs.
To get started with TensorSpace.js, first install it via npm or yarn:
npm install tensorspace
# or
yarn add tensorspace
Then follow the TensorSpace.js HelloWorld documentation or follow this example using CodePen.
The most recent TensorSpace.js 0.2 release adds numerous features and bug improvements. While the project has not yet reached a stable version, it provides a promising collection of useful machine learning data visualizations.
TensorSpace.js is open source software available under the Apache 2 license. Contributions and feedback are encouraged via the TensorSpace.js GitHub project and should follow the TensorSpace.js contribution guidelines.