18- Validation Set Approach To Solving Over-fitting Problem

By using a validation set approach, we can divide our entire data into training and validation sets. The validation set will later be used to determine if we are over-fitting or not!
------------------------------------------------------------
00:00 Recap of previous video and what to come
00:30 Pre-requisites to creating a validation set
01:50 Importing the train_test_split package from Sklearn
03:05 Making changes to the Brain MRI dataset class
08:21 Wrapup and what is to come in the next video!
------------------------------------------------------------
In this video we will understand the concept of validation set in machine learning and see how we can SPLIT THE DATA INTO TRAIN SET AND VALIDATION SET! In order to understand the intuition behind train vs. validation vs. test set division, you should know that the test set is assumed to be ALWAYS hidden from the machine learning model! So, the validation set is some sort of a representative of the test set! The performance of the model on the validation set is assumed to be a decent proxy for the performance of the model on the unseen test set. This is indeed the first step towards one of the MOST IMPORTANT machine learning fundamentals: Cross Validation.

Using a cross validation technique such as 10-fold cross validation, one can solve the issue of underfitting vs. overfitting! By choosing the right set of hyperparameters, the machine learning model will have just the appropriate model capacity to NOT be too simple to underfit the data and NOT too complex either to memorize the training data and overfit!

BUT What is OVERFITTING in Machine Learning? Overfitting is a notorious problem, where a machine learning model is trained so well on the training data, that all it learns is the EXACT values of the training data in the training set! As a result, the model ends up JUST memorizing the training set, which leads to very poor generalization to the unseen data, for example, the test data!

As a result, if it CRUCIAL for you to understand how you can create a validation set from your data!

⚡Join the MLDawn Discord Server: https://discord.gg/u9ZKm7Sm

? GitHub: https://github.com/MLDawn/MLDawn-Projects/tree/main/Pytorch/Brain-Tumor-Detector

? Website: https://www.mldawn.com/

? Twitter: https://twitter.com/MLDawn2018

? Linked In: https://www.linkedin.com/in/mehran-bazargani-14b352176/

Keep up the good work and good luck! ??