ayulockin / SageMakerStudio

A repo showcasing SMSL and W&B

Using Wandb and SageMaker Studio

This repo is acompanied of a blog post in AWS and on fully-connected

Install

This repo can be configured using the pip: requirements.txt or the conda: environment.yml. SageMaker Studio can be configured using both.

Running the experiments

• 01_data_processing.ipynb presents how to log the CamVid dataset and how to do initial EDA.
• 02_semantic_segmentation.ipynb Train a model to performn semantic segmentation on the logged dataset.
• train.py is a refactor of the training notebook into a script to make use of sweeps on the terminal.

Sweep

We will perform an Hyper Parameter Sweep using the sweep.yaml configuration file. To run the sweep with defaults values, use:

$ wandb sweep sweep.yaml

and then launch an agent using the suggested command line output of command above

$ wandb agent <sweep_id>

Note: this repo uses ml_collections to configure parameters. You can check the config.py file to inspect defaults.