# Generate iteration after iteration
result = self.test_input[:250].clone()
+ # Erase all the content but that of the first iteration
result[:, self.it_len :] = -1
+ # Set the lookahead_reward of the firs to UNKNOWN
result[:, self.index_lookahead_reward] = escape.lookahead_reward2code(2)
+
t = torch.arange(result.size(1), device=result.device)[None, :]
for u in tqdm.tqdm(