NotFoundError when running train_qgen_reinforce.py

[n-askarian]

Hi,
In train_qgen_reinforce.py code, when I try to restore qgen in line 163 I am given the following error:
NotFoundError (see above for traceback): Key qgen/rl_baseline/baseline_hidden/W not found in checkpoint
[[Node: save/RestoreV2_4 = RestoreV2[dtypes=[DT_FLOAT], _device="/job:localhost/replica:0/task:0/cpu:0"](_arg_save/Const_0_0, save/RestoreV2_4/tensor_names, save/RestoreV2_4/shape_and_slices)]]

The reason is that qgen_saver which is defined in line 100 expects to restore the variables of QGenNetworkLSTM with policy_gradient=True while the model I want to restore is trained without policy_gradient so it does not have rl_baseline variables.

Could you please help me how to resolve this bug?

Thanks,

[harm-devries]

You can uncomment the part in l.99 of train_qgen_reinforce.py so that you exclude the parameters of the baseline. Then your model should load fine. Note that you won't save the parameters of the baseline but I don't think that's an issue.

@fstrub95 any idea why the RL baseline is not excluded in the qgen saver?

[fstrub95]

I do not remember exactly, yet this commit should fix the issue:
d44caf6