Splitting data to training and testing without validation
3 views (last 30 days)
Show older comments
Dear all,
my dataset has only 20 sample per class and i don't apply any augmentation in order to build a few shot learning model,
i ask is the results wil be correct ? if i test the data to train and test without validation in orde to i can calculate the generalization error
as i know the generalization error is the disability of the model to recognize new unseen data in training
0 Comments
Answers (1)
Krishna
on 5 Oct 2023
Hello Nagwa,
It seems like you're asking whether not using validation error for a few short learning models would lead to a larger generalization error.
The validation error is crucial for assessing the model's performance during training and tuning hyperparameters. The decision to split the data when working with limited data and using short learning models depends on the specific model you are using. In general, having a validation dataset helps prevent overfitting in the model. However, if you already have a small dataset, overfitting might not be a significant concern, so not splitting the data into a validation dataset could be acceptable.
Nevertheless, it is recommended to follow best practices, such as trying cross-validation and different scenarios, to determine what works best for your dataset and the specific few short learning algorithm you are using.
Hope it helps,
Krishna.
0 Comments
See Also
Categories
Find more on Deep Learning Toolbox in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!