How to make predications

[ns-hxue]

Can you teach me how to make predications after model is trained? Does it come with build in method like .predict()?

[ThilinaRajapakse]

Please use Simple Transformers as this repo is now outdated.

To answer your question, this repo does not have such a built-in function whereas Simple Transformers does. If you have already trained a model, you can use it with Simple Transformers as well.

With this repo, you would have to make predictions manually following the same procedure as for evaluation.

[ns-hxue]

Thanks for the advice. If so, how can I save a model from simple transformers and export the model for real-world usage

[ThilinaRajapakse]

from simpletransformers.classification import ClassificationModel


model = ClassificationModel('roberta', '<path_to_directory_containing_trained_model>')
model.predict([<list_of_text>])

[ns-hxue]

Thank you very much! Just want to confirm, if showes you can save the model using train_model method?

After the model being save, how am I supposed to to load the model? Is there a build in method for that, as I didn't see it on the tutorial

[Mahhos]

by '<path_to_directory_containing_trained_model> in model = ClassificationModel('roberta', '<path_to_directory_containing_trained_model>') you mean we shoud just put "output" folder that is generated during training as the directory? I trained the bert model using run_model.py and now need to make preditions for unlabeled dataset I have.
2. We should put these lines of code at the end after the evaluation. right?
I would highly appreciate if you advise me on this.

[ThilinaRajapakse]

This answer was regarding the Simple Transformers library.

I am not sure whether you are referring to Simple Transformers or this repo. Please open an issue on Simple Transformers if it is related to that.

[Mahhos]

I trained the bert model using "run_model.py" in this repository and now need to make predictions for an unlabeled dataset I have. I somehow got confused about how I can make predictions. Would you please advise me on that? Thank you.