In this section we will dive in-depth into hyperparameter tuning which is the natural expansion of experiment tracking.
When you train a deep-learning model for the first time, it's very likely that its performances are not optimal.
That's because the training of a model depends on a lot of parameters (also called hyperparameters) that you will have to tune in order to find the best combination that optimize the performances of your model.
The first step needed to understand why your model performs the way it does is experiment tracking and if you haven't heard of it or don't know how to do it using Picsellia, we suggest you to follow the following tutorials 👇
From now, we will assume that you know how to track your experiments properly, now let's get to the fun part : Hyperparameter Tuning 🎉
At Picsellia, we have developed and engine that allows you to perform hyperparameter tuning easily, we call it Scans.
To learn the basics of Scans and what you can do with it, follow the next link 👇
If you want to run Scans on your local machine or your servers, please check our CLI below 👇
The only thing that you have to do before running Scans is to write its configuration so you can control what will happen very precisely, to see the detailed documentation for the Scan's config 👇
Finally, now that everything should be setup, let's launch the Scans 🚀👇