WeightVis

Visualize a trained neural network weights from different libraries

Table of Contents

• About
• Intallation Started
• TODO
• Contributing
• Authors
• Acknowledgements

About

• Load neural network weights, WeightVis will automatically visualize the neural network weights !
• For now the library works with only for fully connected layers
• Supported Libraries : Pytorch, Sklearn and SimplyNet(https://github.com/frknayk/SimplyNet)
• Tensorflow will be integrated soon !

Usage

• Simplest example with pytorch

from Visualizer.Brain import Brain
from Libraries.Enums import NNLibs as Libs

# Initate visualizer
brain = Brain(nn_lib=Libs.Torch)

# Visualize neural network
brain.visualize("path_your_pytorch_model", load_from_path=True)

• Output

Installation

• pip3 install -e .

TODO

1. Tensorflow is incoming !
2. Extend library to CNNs
3. Arise warning when neura-network size exceeds limits of library
4. Long term TODO : 3D visualisation

Contribution

• Please send an email to [email protected] or [email protected] for contribution or any feedback

✍️ Authors

• @AbdKaan - Idea & developer
• @frknayk - Idea & developer

🎉 Acknowledgements

• The script was first moved from a github user's [@craffel] "draw_neural_net.py" script.