Loading whole model

[eschlager]

1. Feature description

How to load a trained model, without initialising the class beforehand manually?
Since save_model saves the whole model and not only model.state_dict(), the model setup is saved I guess - but I don't know how to load it?

The tutorials only show loading the model after it was initalised previously already.

2. Motivation

After initialising SAITS model, the trained weights can be loaded using saits.load_model(dir).
However, to initialise the model correctly, you need to know n_steps, n_features, etc. before loading the model.

3. Your contribution

I tired loading the model using saits = torch.load(directory).

This loads the module _SAITS.

Running imputation then gives the error AttributeError: '_SAITS' object has no attribute 'impute'.

[WenjieDu]

Please read the doc https://docs.pypots.com/en/latest/pypots.imputation.html#pypots.imputation.saits.SAITS.load_model. Each model in PyPOTS has both functions save_model() and load_model(). After your training, you will find the trained model file saved by PyPOTS. To reload it, you need to initialize a model with the same model hyperparameters, then run model.load_model(path_to_your_saved_model_file).

[WenjieDu]

BTW, if you think SAITS and PyPOTS are useful to your work, please star🌟 them on GitHub and cite them properly in your research 😊 Thank you, Elke! This means a lot to our open-science work.

[github-actions]

This issue had no activity for 14 days. It will be closed in 1 week unless there is some new activity. Is this issue already resolved?