Any guideline on load a pretrained Pytorch NLP model to evaluate new dataset

[zorozy]

❓ Questions and Help

Description Hi,

I am very new to NLP and would like to ask a very stupid question.

After I trained one NLP model, like GRU, what should I save and how could I reload the model again later on a new test dataset?

Besides, what should I do to record the prediction result for each input in the original sequence?

I am so sorry for this baby question.

[pratikkejriwal]

don't know if you are still looking for an answer, but here are few steps that you can follow:

• save the model state

  torch.save(model.state_dict(), '<file_name>')

• recreate the same model structure if you are evaluating in a different program or you can load the state dict to same model in case you just want to use the same program, at a later time
• load the saved model parameters

  new_model.load_state_dict(torch.load('<file_name>'))

as for recording the prediction result of each input (I am assuming you want results for each epoch), you can store those data inside a list:

epoch_data.append([epoch, loss.data.item(), accuracy])

if you want, you can refer to my repository, for further examples/information.