during lecture 2, we used the validation set to see how well a model performs on data that it hasn’t trained on, but so goes for the test set. The only difference is that we used the validation set during training and the test set afterwards, so what is the use of the validation set here?
The validation is used to grasp an idea how well the model performs.
If you would track metrics only on training dataset, you may get a wrong idea how well the learning proceeds. Using validation set allows to check how it performs on never seen data.
The test set is about testing after the training is done. It’s purpose is similar to validation dataset, but it’s evaluated only once (so it might be a bigger one).
Thank you, another quick question. Why is a logistic model considered a linear model even though the latter is used for predicting continuous values and logistic is used for classification
It’s because the outputs are result of a linear transformation of the inputs.
In other words: the model tries to classify something, but it does so by using a linear transformation (in this case, a sum of inputs multiplied by weights).
The output isn’t actually linear, but it is usually a result of some function, which transforms a value that is an outcome of linear operation.