TensorX is a minimalistic utility library to build neural network models in TensorFlow
Project description
TensorX is an free and open-source minimalistic high-level library to build, train, and run neural network models in TensorFlow. This library aims to provide simple, extensible, and modular components, without unnecessary levels of abstraction or verbose.
You can read the documentation on tensorx.readthedocs.io.
Development
Both this library and its documentation are a work in progress, any input is welcome. You can follow the project on Github and read its documentation on ReadTheDocs. I’m focusing on features I myself use in my research, so I’ll add components as I need them. If more people get interested in the project, I’ll create some contribution guidelines.
Installation
TensorX is compatible with Python 3.5+. You can install it directly from this repository using pip
as follows:
sudo pip3 install --upgrade git+https://github.com/davidenunes/tensorx.git
This will install the tensorx
module along with all the required dependencies. If you wish to install a particular version of
tensorflow
(or build it from source) you can either install it first, or install tensorx
with the --no-dependencies
switch and install tensorflow
afterwards. Instructions on how to install TensorFlow
can be found here.
virtualenv: much like tensorflow, virtualenv is recommended to install tensorx in its own environment to avoid interfering with system-wide packages.
Getting Started
Coming soon.
import tensorx as tx
Philosophy
-
Consistent API: simple intuitive API focused on modular neural networks with multiple layers.
-
Pragmatic Code: verbose-free code is more readable, reproducible, and easier to debug and experiment with. Make it easy to use for common use cases.
-
Transparency: the main goal is not to replace the use of TensorFlow or hide it behind abstractions, but to complement it with easy-to-use modular API to create and manipulate tensors.
-
Focus: this is not a library to create every single "Deep Learning" model one might read about. Its about taking advantage of TensorFlow flexibility while compensating for some of its shortcomings.
Author
- Davide Nunes: get in touch @davidelnunes
License
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.