If you like what we're working on, please  star us on GitHub. This enables us to continue to give back to the community.
DEEPCHECKS GLOSSARY

Model Tuning

Hyperparameters are external controls that impact how the model works, similar to how flight instruments influence how an airplane flies. These settings are managed by the user and are not part of the model. They can impact how an algorithm is taught as well as the final model’s structure.

Although past expertise with the model and data may assist, determining the optimal parameters empirically is tough. Manually searching for the ideal hyperparameters would take a long time and consume a significant amount of computational resources. This is why, to get the best settings, automated Hyperparameter tweaking is performed.

Hyperparameter optimization is another term for model tuning. The training process is controlled by hyperparameters, which are variables. During a Model training job, these are configuration variables that do not change. Model tweaking gives ideal settings for hyperparameters, increasing the predicted accuracy of your model.

Each model has its own set of Hyperparameters, some of which are unique to it and others that are shared by a group of algorithms. Maximum leaf nodes are hyperparameters in XG boost, whereas several layers and hidden width are hyperparameters in Neural Networks.

When adjusting hyperparameters to check if the model improves, keep the following in mind:

  • Which hyperparameters have the most impact on your model?
  • Which values should you choose?
  • How many hyperparameter combinations should you try?

Tuning practices

Hyperparameter tuning, also known as optimization, may be a time-consuming process. Best practices can be used to control resource needs and increase optimization.

  • There are a certain number of hyperparameters. While SageMaker only allows you to search 20 hyperparameters, it is recommended that you search for more. This is because the amount of hyperparameters in the Search Space increases the computational complexity.
  • Ranges of hyperparameters Limiting the range of hyperparameters to be searched will yield better results. This is where past optimization knowledge with a certain sort of data and technique might be useful. The size of the Search Space is controlled by limiting the range.
  • For hyperparameters, log scales are used. It will first assume that a variable is linearly adjusted, and will only process results as logged scaled once the variable has been found to be logarithmic. To speed up processing, convert log-scaled data to linear-scaled variables.
  • Running a large number of training tasks at the same time will speed up the optimization process, but sequential processing will yield superior results. This is because each training task finished produces knowledge that may be used to better the following training job. There is far less potential to share this information with the following employment with concurrent training positions. As a result, concurrency is a compromise between speed and quality.
  • Multiple instances are being used. There is a similar communication issue when executing training jobs on several instances as when running jobs concurrently. You must ensure that the proper objective metric is conveyed and implemented.
  • Use Bayesian search to find what you’re looking for. Bayesian search is a more effective, less expensive, and quicker method of tuning hyperparameters. Random search often necessitates 10 times the number of jobs that Bayesian search necessitates.

External controls called hyperparameters are utilized to regulate how the model trains and runs. Hyperparameters, on the other hand, may be manually regulated and optimized by trial and error.

Open source package for ml validation

Build Test Suites for ML Models & Data with Deepchecks

Get StartedOur GithubOur Github
×

Event
Identifying and Preventing Key ML PitfallsDec 5th, 2022    06:00 PM PST

Days
:
Hours
:
Minutes
:
Seconds
Register NowRegister Now