https://github.com/srowen/autoenc-neurons

Simple example of Keras, TensorFlow and autoencoders applied to text
https://github.com/srowen/autoenc-neurons

autoencoder gpu gutenberg jane-austen keras neurons tensorflow

Last synced: 2 months ago
JSON representation

Simple example of Keras, TensorFlow and autoencoders applied to text

• Host: GitHub
• URL: https://github.com/srowen/autoenc-neurons
• Owner: srowen
• License: apache-2.0
• Archived: true
• Created: 2017-07-23T10:46:02.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2018-05-30T23:16:05.000Z (over 7 years ago)
• Last Synced: 2025-01-28T03:35:10.053Z (11 months ago)
• Topics: autoencoder, gpu, gutenberg, jane-austen, keras, neurons, tensorflow
• Language: Python
• Homepage:
• Size: 1.65 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# autoenc-neurons (work in progress)

An experiment with convolutional autoencoders as applied to text, and exploring what the

'neurons' in the convolutional layers respond to.

## Running

Note that if you do not have GPUs, you will need to edit the script to install `tensorflow` 

instead of `tensorflow-gpu`, and change `gpu_count` to 0.

```bash

python3 autoenc-neurons.py

```

This will start model training and then produce output, at last, like this:

![](docs/img/neuron_output.png)

The output contains several lines from the text, repeated several times. In each

stanza, the letters are shaded according to the activation that it causes a particular neuron

(darker means more active). This is repeated for several neurons.

## References

The source text is The Complete Project Gutenberg Works of Jane Austen:                       

http://www.gutenberg.org/files/31100/31100.txt

For more on convolutional autoencoders in Keras, see:

https://blog.keras.io/building-autoencoders-in-keras.html