mila_tools
aims to reduce the overhead to deploy experiments on mila clusters.
experiment_buddy
aims to reduce the overhead to deploy experiments on servers.
It's important to reduce cognitive overload for the researcher measured as seconds-to-first-tensorboard datapoint
Right now it's responsabilities cover:
- Deployment on servers
- Handling of Sweeps
- Tracking of hyperparameters
- Code versioning
- Notifications
- Wandb integration.
Example: (Updated)
python3 -m pip install virtualenv
python3 -m virtualenv venv --python=python3.8
source venv/bin/activate
- Add cluster private-key to:
https://github.com/settings/keys
if you dont know this yet, check this out here pip install wandb && wandb init
there are two ways to set up wandb:- set up wandb init from the cmd
- or set export WANDB_API_KEY = your wandb key which can be found here: https://wandb.ai/settings
python -m pip install jax jaxlib
# Local requirementspython -m examples.mnist_classifier
More details on experiment-buddy:
- experiment-buddy will commit to a branch called "experiment-buddy" to the first git repo found in the reverse os walk tree
- Supports: Unix based OS