jbegaint / hualos

Keras Total Visualization project

Hualos - Keras Total Visualization project

For now, this is a simple demo where a Flask server exposes an API to publish and consume events in the form of JSON objects. The Hualos callback RemoteMonitor can publish events to the server, and the Hualos landing page listens to the server and displays incoming data on a c3.js graph.

Installation

• git clone https://github.com/gabrieldemarmiesse/hualos
• pip install hualos

Usage:

• start the server: hualos-serve
• open the landing page: http://localhost:9000/
• launch a Keras experiment with the RemoteMonitor callback:

from hualos.HualosCallback import RemoteMonitor
remote = RemoteMonitor(root='http://localhost:9000')

model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch, validation_data=(X_test, Y_test), callbacks=[remote])

• the available metrics of the experiment will start being graphed in real time on the landing page.

Help

• hualos-serve --help

Dependencies:

• Python:

  • Flask
  • gevent

• JS (included in the repo):

  • d3.js
  • c3.js