Hello, I have trained a neural network using pointwise constraint, and now I need to add new training data, what is the correct way to train the pretrained network?
You should be able to have Modulus load a model and train on it again with another script. For loading a already trained model there are some examples of this in our examples such as the Three Fin where the flow model is already trained (in this case it not optimized here).
In general, Modulus will attempt to load the model checkpoint by default on initialization. So if you’ve got the checkpoint already in the correct directory theres no extra work needed. This will call the
_load_model() method. You can also load it manually, in your script using
model.load(path, mapping), see example in the trainer.
You can then create a new constraint for the new data and train.