tensorstate 0.2.4

Tools for analyzing neural network architecture.

TensorState - Neural Network Efficiency Tools

TensorState is a toolbox to capture neural network information to better understand how information flows through the network. The core of the toolbox is the ability to capture and analyze neural layer state space, which logs unique firing states of neural network layers. This repository implements and extends the work by Schaub and Hotaling in their paper, Assessing Intelligence in Artificial Neural Networks.

Installation

Precompiled wheels exist for Windows, Linux, and MacOS for Python 3.6-3.8. No special installation instructions are required in most cases:

pip install pip --upgrade

pip install TensorState

If the wheels don't download or you run into an error, try installing the pre-requisites for compiling before installing with pip.

pip install pip --upgrade

pip install numpy==1.19.2 Cython==3.0a1

pip install TensorState

Documentation

https://tensorstate.readthedocs.io/en/latest/