Load/restore model without rebuilding data and network object?

[Nhatkhang]

Hi all,
I am encountering a bottle neck when restore a pre-trained model for a new set of data (the data only have input for prediction). However, when I tried to restored the model from a pre-trained model, its requires data object and net object.
The data object is something looks like: data = CartesianProd(x_train, y_train, x_test, y_test). Obviously, there is no y_train, x_test, y_test available here. How can I resolve this issue without loading the training data to create the same data object, and is there a way to prevent this loading every time a new set of data is inputed?

Thank you in advance.

[vl-dud]

I use fake data for such case:

# generate CartesianProd that wraps numpy arrays
# array sizes depend on net input/output
data = generate_fake_data(net)

model = dde.Model(data, net)
model.compile()
model.restore(checkpoint_path)

This is just my solution, not sure if this is the best way.
@lululxvi Could you please comment on this?

[Nhatkhang]

Is the generate_fake_data avaible in deepxde library or just a function to create similar data shape?

[vl-dud]

Just my custom function

[lululxvi]

Yes, just use some fake data, as the data will not be used if you don't train the model.