I do. This is why I wrote ML-Logger and a ML-Dash, two open-source distributed logging and visualization library for you, and your University collaborators who want to work on the same code!
To get started:
pip install ml-logger ml-dash
python -m ml_dash.app
/Users/david/berkeley/packages/ml_logger/ml-dash-server/ml_dash/client-dist
You can now view ml-dash client in the browser.
Local: 'http://localhost:3001/
To update to the newer version, do
~> pip install --upgrade ml-dash
[2019-06-05 20:56:33 -0700] [46329] [INFO] Going Fast @ 'http://127.0.0.1:3001
[2019-06-05 20:56:33 -0700] [46329] [INFO] Starting worker [46329]
supports two way communication between the job and the instrumentation server. Want to run some analysis on a model trained last week? No problem! just do
logger.load_module(<NNModule>, key="username/project/run-id/models/blah.pkl").
Did I mention that ML-Logger is uber-fast? We make the IO requests asynchronously, so that you main code doesn’t slow down. We also support local metrics cache, so that you only send the summary of the metrics :).
Oh, for logging videos, we first compress the frame tensor 200x. And we support live plotting with Matplotlib!
A Visualization Dashboard designed from ground up, to replace Tensorboard and Visdom.
Pictures are worth a thousand words–see below!
- Setup
- Old Tutorial [delete soon]
- logging Data
- Detailed API Doc
- Contribute
- Develop the ml dash
- How to generate GraphQL schema