Should I use the dev set to iteratively train the model or just use it to detect the error rate？The answer might be wrong，but I want to know why.Thank you
Hi, @1004271927!
Dev set is normally used to check how well your trained model performs with unseen data. Guided with these results, you can adjust your hyperparameters to get better results. Nevertheless, as you tweak this values to maximize the metrics on dev set, you might end up overfitting this hyperparameters on the dev set. That’s the reason behind having an additional test set.
You may find this post useful as well.
3 Likes