How to initialize the deepxde model with a trained neural network?

[mynanshan]

Suppose that I am doing an inverse problem. I want to train the deepxde model with some rough data first, and then continue updating the trained neural network with a new dataset.

Then the question comes down to, how can I use the outcome as the initialization in the second training stage, or equivalently, change the dataset in the middle.

In the current deepxde framework with model = dde.Model(data, net), it seems that we have to create a new model object whenever switching to a new dataset.

[vl-dud]

model = dde.Model(first_data, net)
model.compile("adam", lr=0.001)
model.train()
model.data = second_data
model.compile("adam", lr=0.001)
model.train()