For my timeseries problem it seems obvious to use teacher forcing. For example in the case of predicting the new timestep in a real life scenario, I do have access to all the ground truths for all previous steps so using this information would paint a more accurate picture.

But, in my current implementation of a LSTM, I don't see an obvious way to implement teacher forcing. I have an lstm with hidden dimension of 20. After the lstm I apply a linear layer transforming the data to one dimensional. As the hidden dimension is 20, I don't see how I can input the one dimensional ground truth into the model. Is there something obvious I'm missing, or can someone point me to some related literature? Thanks!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.