CPC weights not part of optimizer

[kinalmehta]

Thank you for a clean implementation of CPC in RL.

One of the query I have is as below:
As you can see in the blow link, the optimizer is only given the weights of the actor-critic model and not the Wk_state. Is this normal or something is missing here?

Contrastive-Predictive-Coding-in-RL/core/algorithms/a2c_acktr.py

Lines 27 to 31 in 1d4c102

	if acktr:
	self.optimizer = KFACOptimizer(actor_critic)
	else:
	self.optimizer = optim.RMSprop(
	actor_critic.parameters(), lr, eps=eps, alpha=alpha)

Contrastive-Predictive-Coding-in-RL/core/algorithms/cpc_a2c.py

Lines 28 to 31 in 1d4c102

	self.Wk_state = nn.ModuleList([nn.Linear(hidden_dim, hidden_dim, bias=False) for i in range(num_steps)])
	self.Wk_state_action = nn.ModuleList([nn.Linear(hidden_dim, hidden_dim, bias=False) for i in range(num_steps)])
	self.Wk_state = self.Wk_state.to(device)
	self.Wk_state_action = self.Wk_state_action.to(device)

Thanks
Kinal