What is wandb?
wandb is an acronym for Weights & Biases, a tool that helps keep track of your machine learning projects. It can automatically record hyperparameters and output metrics during model training, then visualize and compare the results, and quickly share the results with colleagues.
Through wandb, you can bring a powerful interactive visual debugging experience to your machine learning project, automatically record the icons in the Python script, and display its results on the web dashboard in real time, such as loss function, accuracy rate, and recall rate , which allows you to complete the visualization of machine learning projects in the shortest possible time.
In summary, wandb has 4 core functions:
Kanban: Track the training process and give visual results
Report: Save and share some details and valuable information during the training process
Tuning: Use hyperparameter tuning to optimize your trained model
Tools: Dataset and model versioning
That is to say , Wandb is not simply a data visualization tool. It has more powerful model and data version management. Additionally, it is possible to fine-tune the models you train.
Another highlight of wandb is its strong compatibility. It can be used in conjunction with Jupyter, TensorFlow, Pytorch, Keras, Scikit, fast.ai, LightGBM, and XGBoost.
Therefore, it not only saves you time and energy, but also brings a qualitative change in your results
How to use wandb?
1. Visit the wandb website, the wandb website is wandb.ai, but to access this website, you need to bypass the wall, so I won’t repeat how to bypass the wall here
2. Register a github account, and then use the github account to register a wandb account
3. Create a new project, click Create new project
4. After creating a new project, open pycharm to see which virtual environment wandb wants to install in
I want to install it in the lvnan virtual environment, then enter this virtual environment, enter pip install wandb, and follow it
5. The last step
Click Settings, then pull down the page to find the API Key
Then enter the virtual environment where wandb is installed again, enter
This own API Key is copied from the API Keys in the picture above, and you're done.
6. Results display
At this time, we can train our own data set. During training, wandb's official website will display that the program is running, and then dynamically generate useful curves for us. Like this, there are many other functions, you can explore by yourself Bar!
Like or pay attention to the little hands who make a fortune, and we will continue to work hard to write good articles in the future!