Wandb + TensorFlow

About The Project

It is a simple example demonstrates how to use the Wandb API to monitor and analyze your Machine Learning projects.

Wandb page of this project: https://wandb.ai/anon234523452346/CIFAR10

With free Wandb account you will have a 100 GB of cloud storage for your logs, models, artifacts, and other data.

It also uses Docker and Docker Compose. And runs on GPU and CPU as well.

Getting Started

1. First thing first, you should have a Wandb account to monitor your experiments.
2. You need a Wandb API key. You can get one from Wandb account settings.

Dependencies

1. Python

tensorflow==2.9.1
tensorflow_datasets==4.6.0
keras-cv==0.2.6
PyYAML>=6.0
wandb>=0.12.17
matplotlib>=3.5.2

2. wandb account for tracking your experiments.

• Create here: https://wandb.ai/signup

Installation

1. Clone this repo

   git clone https://github.com/Alex-Kopylov/wandb-tensorflow-example.git

2. Setup your environment. You have several options:

Docker is preferable for further integration in complex CI\/CD pipelines.

  docker compose build
  docker compose up

You can use Conda or default Python virtual environment.

  conda create -n wandb-tensorflow python=3.8
  conda activate wandb-tensorflow
  pip install -r requirements.txt

Hints

1. For dry run:

   $ wandb disabled

   $ export WANDB_MODE=disabled

   wandb.init(mode="disabled")