L-BFGS freezes with 'tensorflow'

[pescap]

Hi,

deepxde/examples/pinn_forward/Poisson_Lshape.py

Line 25 in 4cd2178

losshistory, train_state = model.train()

freezes when using tensorflow as a backend, in CPU and GPU modes.

Example with 5000 ADAM epochs:

'train' took 58.066181 s

Compiling model...
'compile' took 0.019574 s

Training model...

Step      Train loss              Test loss               Test metric
5000      [6.19e-04, 5.82e-04]    [6.10e-04, 5.82e-04]    [] 

and then, nothing.

[lululxvi]

It is still running. Just wait.

[ShaikhaTheGreen]

I have another question related to L-BFGS. In some of my runs it only performs one step of iterations (1000 epochs) and then the training concludes. Is this normal?

[lululxvi]

See https://deepxde.readthedocs.io/en/latest/modules/deepxde.optimizers.html#deepxde.optimizers.config.set_LBFGS_options