Scan Philosophy

Overview

17061706

In machine learning, hyperparameter optimization or tuning is the problem of choosing a set of optimal hyperparameters for a learning algorithm. A hyperparameter is a parameter whose value is used to control the learning process. By contrast, the values of other parameters (typically node weights) are learned.

The same kind of machine learning model can require different constraints, weights, or learning rates to generalize different data patterns. These measures are called hyperparameters and have to be tuned so that the model can optimally solve the machine learning problem.

Hyperparameter optimization finds a tuple of hyperparameters that yields an optimal model which minimizes a predefined loss function on given independent data. The objective function takes a tuple of hyperparameters and returns the associated loss.

It can be kind of scary, but Picsellia can help you find the best hyperparameters for your models ๐Ÿ”ฅ

TL; DR;

Picsellia platform can help you find the best hyperparameters for your ML models, you just need to create a scan using our Python SDK, define the metric to optimize and the strategy to use ( grid, bayesian, etc. ), and launch it on your local machines or on Cloud GPU if you have an enterprise Picsellia account.

Jumpstart your hyperparameters tuning by following these steps :

  • Initialize a Hyperparameter Scan

  • Make a training Loop

  • Hit run and go get some quality sleep, it's important!

What's a Scan ?

Picsellia platform allows you to create Scans, which refers to a set of runs designed to test hyperparameters and find the best ones for your ML models.

Basically, you need to set all the hyperparameters that you want to tune and the parameters to optimize, for example, the evaluation loss

The platform will then generate all the exploration runs to launch and will act as an Oracle to send you the next training after yours is over.

Benefits:

  • Distribute your hyperparameters search across all your organization laptop or servers
  • Benefits from Optuna Bayesian Search and early stopping in order to ensure minimum time to converge to optimum
  • Control all your machines from Picsellia Platform

Focus on code

You can pass the script to be run, your requirements, the parameters to explore, we do the rest ๐Ÿš€
Once everything is set up, you can let your machines and Picsellia do the rest,

Leverage all your company laptops or server to speed up the process

Creating a machine swarm has never been so easy, just launch the training loop on all your machines, then all the available runs will be sent to all the available machines, insuring you to exploit all the processing power you have to converge the fastest way possible!


Did this page help you?