Exploration tool for your NeuralNetwork
TuringNetwork is built to make model interpretation easy. Rather than just being a visualization tool we wanted to make a playground for researchers and developers to debug the neural network and playe around with the graph. TuringNetwork currently supports only PyTorch model visualization but we are tyring to make it compatible with GLUON and TensorFlow eager.
TuringNetwork has very intuitive and non-exhaustive list of APIs available. The TuringNetwork object creation expects the user to pass the neural net instance
>>> import TuringNetwork >>> net = NeuralNet() >>> tnet = TuringNetwork(net)
Calling the functional APIs saves the data to the TuringNetwork home folder which is by default
TuringNetwork installation creates the command line interface which also to call the TuringNetwork server. TuringNetwork server collects the data from TuringNetwork home folder and serves in the browser.
TuringNetwork uses sanic as backend server. Sanic is dependant on python
await and hence won't work with versions < python 3.5
If your code base is python 3.5+,
Everything works fine by installing TuringNetwork from pip. This will give you access to importable TuringNetwork package and command line interface that instantiates the Sanic server
pip install turingnetwork
If your code base is using python version < 3.5,
In this case, you are in bit trouble. We have created a separate TuringNetwork client for you to import in your script but TuringNetwork server still have to be with python3.5+, that means, you'll end up installing
turingnetwork-client in your environment that uses older version of python and you need to create an environement with new version of python for running the TuringNetwork server. We really wanted to help you and hence we had created docker image also for you. Step for people who uses older version of python will be updated once we have a beta release.
Full Documentation is available in readthedocs