Tensorflow

From Robin

Jump to: navigation, search

800px

TensorFlow is an open source library for defining and evaluating computational graphs. It’s great for developing Artificial Neural Networks and includes many useful components (e.g., LSTM and various RNN cell designs) and optimisation tools (e.g., Adam).

Tensorflow is most useful if you want to work at the operation level of machine learning development, e.g., defining special kinds of output layers. If you want to work with existing layers, it may be better to use a higher-level library such as Keras.

Tensorflow is still new (1.0 came out in February 2017) and can be a bit of a moving target (particularly the contrib parts), but it includes some great tools such as Tensorboard for visualisation and is good at computing things on GPUs.

Contents

Installing:

Lots of good advice on the TensorFlow site: https://www.tensorflow.org/install/ but more or less:

  1. Install python of choice
  2. pip install tensorflow
  3.  ??
  4. $$!

Resources:

Here’s some useful resources to get started:

Courses

Books

Blogs

Examples:

Personal tools
Front page