Train-Dev error

How do we get an estimate of the train-dev error if we aren’t going to train that data ?

Hey @ram,

welcome to the DLS! :slight_smile:

We use a train-dev set in the same way we use a dev or a test set. To calculate an error we pass train-dev set data as inputs to our model and evaluate results.

1 Like

Andrew mentioned in the lecture that we are not going to explicitly train the train-dev set. Does he mean that we just do forward propagation on the train-dev set and get an error value from it ?

That’s correct. We train our model on the train set and then cross-validate on the train-dev set.