Saving Lagrange multipliers with NN critic networks

[jamesjingqili]

I wonder if the Lagrange multipliers are saved automatically somewhere when we save the policy as .pt file? If so, how can I reload it? I plan to plot the combination of critic[0] - lambda * critic[1] . Thanks!

[liuzuxin]

Hi @jamesjingqili , in our provided example scripts we only save the policy parameters. If you want to save the Lagrange multipliers along with the policy in the checkpoint, consider revising this checkpoint callback function.

You can also override the policy's get_extra_state functions to save any other things, see here for an example.

You can also access all these training metrics in the trainer here