Enables batch DDPG agents to be trained.

[prabhatnagarajan]

No description provided.

[prabhatnagarajan]

This PR changes two Q-functions: FCBNLateActionSAQFunction and FCLateActionSAQFunction. A simple search in the ChainerRL repository:

https://github.com/chainer/chainerrl/search?q=FCBNLateActionSAQFunction&unscoped_q=FCBNLateActionSAQFunction

https://github.com/chainer/chainerrl/search?q=FCLateActionSAQFunction&unscoped_q=FCLateActionSAQFunction

shows that the DDPG example is the only place these are used. Furthermore, looking at the __init__ functions for those networks, it appears that neither of them need recurrence. Since

chainerrl/chainerrl/links/mlp.py

Line 16 in 15d7cbb

class MLP(chainer.Chain):

and

chainerrl/chainerrl/links/mlp_bn.py

Line 31 in 15d7cbb

class MLPBN(chainer.Chain):

are not recurrent.

[prabhatnagarajan]

[muupan]

👍