Utilites for hyperparameter sweeps, using Weights and Biases. These instructions are a minimial subset of these docs.
To begin a sweep follow these steps:
-
Make sure wandb is installed:
pip install wandb
-
Login to wandb:
wandb login
-
Initialize wandb (optional)
wandb init
-
When asked to choose a project make sure you select
covasim
. If you don't see a project with this name, instead selectCreate New
and name your projectcovasim
. -
From the root of this repo, initialize a sweep.
# Choose the yaml file that corresponds to the search strategy. wandb sweep sweep/sweep-random.yaml
This command will print out a sweep ID. Copy that to use in the next step!
-
Launch agent(s)
wandb agent your-sweep-id
From the docs:
You can run wandb agent on multiple machines or in multiple processes on the same machine, and each agent will poll the central W&B Sweep server for the next set of hyperparameters to run.