Get more out of your ML workflow:

Minimal MLOps everyone should know

This repository is for the GDG Cloud Kolkata workshop on 27th August, 2022. Learn about minimal MLOps required for any ML practitioner. Find the curated list of resources here.

⭐ Slide Deck: http://wandb.me/minimal-mlops-deck

Prerequisite setup

Free W&B Account

If you don't have a free W&B account follow the steps:

• Visit wandb.ai/site.
• Click on sign-up and follow the signup process.
• Login

Authenticate your machine with W&B authorization key

To authenticate any machine to start logging experiments to your W&B account,

• visit wandb.ai/authorize to get the key.
• alternatively, you can visit your settings page to get the key.

Quickstart

Install repository

1. git clone https://github.com/ayulockin/minimal-mlops
2. cd minimal-mlops
3. pip install -e .

Install the dependencies

1. pip install -r requirements.txt

Configuration

Everything is stitched together using configs file. You can find the config for this repo in configs/ dir.

Train without W&B

python train.py --config configs/baseline.py

Train with W&B

python train.py --config configs/baseline.py --wandb

Checkpoint without W&B

python train.py --config configs/baseline.py --log_model

Checkpoint with W&B

python train.py --config configs/baseline.py --log_model --wandb

Log evaluation

python train.py --config configs/baseline.py --wandb --log_eval

Log data

python log_data.py --config configs/baseline.py

I hope you find it useful. If you encounter any issue please raise an issue. :)