Skip to content
Monitor training in Caffe
Branch: master
Clone or download
Pull request Compare This branch is 36 commits ahead of Reportr:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
bin
lib
public
test
.gitignore
API_REFERENCE.md
Gruntfile.js
LICENSE
Procfile
README.md
app.json
bower.json
package.json
preview.png

README.md

Mri-Server

Neural network monitoring

This project is based on Reportr, the open source dashboard. For instructions specific to Reportr, please see the project homepage.

Mri-server constitutes the web-based monitoring portion of Mri. When used together with the Mri-app for Caffe or the Mri-python-client, it allows you to watch the progress of your networks as they train from anywhere, as well as test multiple hyperparameters or architectures at once.

The project is entirely open source and you can host your own Mri-server instance on your own server or Heroku.

Screen Preview

Installation

For installation instructions, see the documentation

Configuration

Reportr is configured using environment variables.

Name Description
PORT Port for running the application, default is 5000
MONGODB_URL Url for the mongoDB database
REDIS_URL (Optional) Url for a redis database when using worker mode
AUTH_USERNAME Username for authentication
AUTH_PASSWORD Password for authentication

Running with Mri-app or Mri-python-client

The Mri clients already know how to talk to the server, and will automatically create reports and visualizations as you train networks. Simply modify the Mri-app configuration file to properly interface with the server as a dispatch. See API_REFERENCE.md for full API specifications.

Scale it

Reportr can easily be scaled on Heroku (and compatibles), use the REDIS_URL to enable a task queue between workers and web processes.

You can’t perform that action at this time.