ModelCheckpoint in L-BFGS: no '.pt' file

[ShengqiangSKR]

I use ‘adam’ as my first optimizer, after 200000 epochs, restore the best model, then i change to 'L-BFGS-B', beacause nan was found during the L-BFGS-B training process, so i add a ModelCheckpoint into it. at the adam stage, the '.pt' file generated, but at the L-BFGS-B stage, no '.pt' file was found. MAYBE i did something wrong in my code, can anyone help? here is me code:

data = dde.data.TimePDE(
geomtime, pde_euler_1d,
[ic_rho, ic_p, ic_u, bc_u, bc_p_l, bc_p_r, bc_rho_l, bc_rho_r,],
num_domain=10000, num_boundary=1000, num_initial=1000)
net = dde.nn.FNN([2] + [80] * 7 + [3], "tanh", "Glorot normal")
model = dde.Model(data, net)
loss_weights = [1, 1, 1, 10, 10, 10, 1, 1, 1, 1, 1]
model.compile("adam", lr=1e-4, loss_weights=loss_weights)
checkpointer = dde.callbacks.ModelCheckpoint("model/model.ckpt", save_better_only=True, period=1000)
losshistory, train_state = model.train(epochs=3000, (callbacks=[checkpointer])
model.restore("./model/model" + ".ckpt-" + str(train_state.best_step) + '.pt', verbose=1)
model.compile("L-BFGS-B", loss_weights=loss_weights)
losshistory, train_state = model.train(callbacks=[checkpointer])
dde.saveplot(losshistory, train_state, issave = True, isplot = True)

[lululxvi]

L-BFGS will not save checkpoints during optimization.

[ShengqiangSKR]

i got nan at the L-BFGS-B optimization, if i can't save the checkpoint, how can i save the model params

[lululxvi]

It will save in the last step.

[ppandit95]

Dear @lululxvi
I am facing a similar issue with "L-BFGS" optimizer as I have some constants which need to be trained in the governing equation but after optimization I am not able to get the values of trained constants.So is there a way to output these values after running the "L-BFGS" optimizer.

Many Thanks
Pushkar

[lululxvi]

Try the latest DeepXDE version.

[ppandit95]

Thanks @lululxvi for suggesting that but will it be possible to update DeepXDE with pip

[ppandit95]

I did manage to update the version of DeepXDE with pip and now the learned variables are stored in a dat file