Saving model with best loss in NeuralPDE

[gradureau]

Hey everyone,

I am trying to use NeuralPDE to solve equations and I’ve been trying to use the QuasiRandomTraining method, with Adam optimizer. Now the learning curve is quite chaotic, so I would wish to save the best model encountered during the training instead of the last. Is there a way to do this?

Thank you for your answers!

[ChrisRackauckas]

@gradureau

I am trying to use NeuralPDE to solve equations and I’ve been trying to use the QuasiRandomTraining method, with Adam optimizer. Now the learning curve is quite chaotic, so I would wish to save the best model encountered during the training instead of the last. Is there a way to do this?

IIRC Optimization already is setup to best save by default. @Vaibhavdixit02 what’s the option on that?

[Vaibhavdixit02]

It’s a kwarg to solve function save_best and is true by default