Teacher MLP Input Dimension Mismatch When Teacher is Non-Recurrent

[Passutte]

In StudentTeacherRecurrent.init, when teacher_recurrent=False, the teacher MLP is initialized with the wrong input dimension. This bug was introduced in #116.

if self.teacher_recurrent:
    self.memory_t = Memory(
        num_teacher_obs, type=rnn_type, num_layers=rnn_num_layers, hidden_size=rnn_hidden_dim
    )
self.teacher = MLP(rnn_hidden_dim, num_actions, teacher_hidden_dims, activation)

Proposed fix:

# teacher
if self.teacher_recurrent:
    self.memory_t = Memory(
        num_teacher_obs, type=rnn_type, num_layers=rnn_num_layers, hidden_size=rnn_hidden_dim
    )
    teacher_input_dim = rnn_hidden_dim
else:
    teacher_input_dim = num_teacher_obs
self.teacher = MLP(teacher_input_dim, num_actions, teacher_hidden

[ClemensSchwarke]

Hey @Passutte, thanks for bringing this up! #136 should fix this :)