PiNNboard is a plugin for Tensorboard to visualize atomic neural networks. It provides a way to define and visualize the activate and weights of atomic neural networks, with special focus on the PiNet neural network.
This visualization is much inspired by the tensorboard playground in terms of both the idea and the style.
PiNNboard currently depends on the nightly version of tensorboard for dynamical plugin support. To install:
# Preferably do this in a virtual env.
pip install tf-nightly
# clone the repository
cd PiNNboard && pip install -e .For visualizing a PiNet, see the example in the notebooks/ folder.
WARNING PiNNboard is currently under active development, and it relies on experimental features of Tensorboard.
- The dynamic plugin feature is only available in the nightly verion of tensorboard, which is expected to move to TF 2.0 soon.
- The fontend part of the code has much to improve, e.g. more options, better UI, etc.