Optimizer as a Class Attribute

[davera-017]

In QNetwork and DuelingQNetwork brains the optimizer is getting initialized at every update step, which makes the training significantly slower:

def update(self, _, pred, target):
        optimizer = torch.optim.Adam(self.parameters(), lr=self.learning_rate)

Maybe a solution will be to convert it into a class attribute. Ej.:

class QNetwork(nn.Module, Brain):
    def __init__(self, env: gym.Env, hidden_layers=[], learning_rate=0.01, alpha=0.001):
        ...
        self.optimizer = torch.optim.Adam(self.parameters(), lr=self.learning_rate)

[aristizabal95]

Good catch! And a pretty simple one to fix. Do you want to create a PR for this, or should I take it?