This is a simple example to showcase Weights and Biases artifacts. As part of this example, we train EfficientNet-B3 on Imagenette Dataset.

For a complete report that explains using W&B for compliance and audit purpose, please refer here.

To prepare the dataset, run the following lines of code in the root folder of this directory:

mkdir data && cd data 
wget https://s3.amazonaws.com/fast-ai-imageclas/imagenette2-160.tgz
tar -xvf imagenette2-160.tgz

Now you should have a data directory in the repository whose folder structure looks like:

data/
└── imagenette2-160
    ├── train
    │   ├── n01440764
    │   ├── n02102040
    │   ├── n02979186
    │   ├── n03000684
    │   ├── n03028079
    │   ├── n03394916
    │   ├── n03417042
    │   ├── n03425413
    │   ├── n03445777
    │   └── n03888257
    └── val
        ├── n01440764
        ├── n02102040
        ├── n02979186
        ├── n03000684
        ├── n03028079
        ├── n03394916
        ├── n03417042
        ├── n03425413
        ├── n03445777
        └── n03888257

From the root of this folder, simply run:

python src/train.py

For this script to work:

1. Please make sure that you have setup AWSCLI
2. Please make sure that you read/write permissions to AWS bucket

After, simply run:

python src/upload_artifact_to_s3.py --project <enter-wandb-project-name> --filename <enter-artifact-name> --alias <enter-alias> --bucket <enter-aws-bucket>