When training fullplace

[PKUTAN]

when I train, it do not seem to coverge

[Rogerity]

Due to the immense action space, placement is actually a difficult RL learning task, so it is common to face such situation. There are several possible reasons for your curve, such as lack of pre-training or overfitting. To make training more stable, tuning hyperparameters such as reward funtion and learning rate may be helpful as well.

[PKUTAN]

Can you show the hyperparameters to train fullplace environment?