wandb.ai


A few times in your life you will experience situations when you’ll see something and immediately know that “that’s it”. My latest that kind of situation was when I opened wandb.ai or the first time.

Let’s start from the beginning.

So, you want to train an AI model. Of course, there would be a lot of experimenting, since there is no silver bullet for all the datasets. And you have a bunch of meta parameters that you have to vary to achieve better results. And you have to write logs over and there. And you have to keep track of your experiments, i.e., what have you changed to produced your latest experiments.

How much of extra time have you spent on documenting and organizing things? The more you advance in your work to optimize a model, the more time you spend on this. Hopefully, you have gathered all the information needed to reproduce the model. Otherwise, that model is gone. And finally, when you spent some time experimenting, you decide that you have to automate the experiments somehow to write a script that will automatically vary hyperparameters and execute experiments, writing them down.

How many lines of code have you created to achieve this?
And finally, you are facing the truth that your experiment will last for a while. Now, you figure out that it would be cool if you have a mechanism for tracking the parameters of the model’s behavior on your mobile phone, wherever you are.

Now you are in trouble. Because you can’t write a code that will allow this to you unless you decide that you programmatically send an email to yourself every 50 steps in your model training. You can’t write it because you are not a professional front-end developer. That’s not the area of your expertise and interest.

If you use wandb.ai:

  • new lines of code will be 15-25
  • you have a well-organized and beautiful view of all your experiments
  • all of the data needed for reproducing the experiment is stored, with the trained model that is produced
  • all the models are stored on their server
  • hyperparameters tuning done automatically when you decide to go with this

Sounds great?
It did to me.
And at the end, all of this comes for free, if you don’t train and share your model together with your teammates. With great online support from their personnel, and Slack community.

Don’t waste your time. Go there and enjoy 🙂
& Have a nice weekend.

Machine Learning

AIdata scienceDeploying AI modelsFeature EngineeringML

Leave a Reply

Your email address will not be published. Required fields are marked *

twelve + 18 =