def __init__(self, opt):
super(LeakGANInstructor, self).__init__(opt)
# generator, discriminator
self.gen = LeakGAN_G(cfg.gen_embed_dim, cfg.gen_hidden_dim, cfg.vocab_size, cfg.max_seq_len,
cfg.padding_idx, cfg.goal_size, cfg.step_size, cfg.CUDA)
self.dis = LeakGAN_D(cfg.dis_embed_dim, cfg.vocab_size, cfg.padding_idx, gpu=cfg.CUDA)
self.init_model()
# optimizer
mana_params, work_params = self.gen.split_params()
mana_opt = optim.Adam(mana_params, lr=cfg.gen_lr)
work_opt = optim.Adam(work_params, lr=cfg.gen_lr)
self.gen_opt = [mana_opt, work_opt]
self.dis_opt = optim.Adam(self.dis.parameters(), lr=cfg.dis_lr)
# Criterion
self.mle_criterion = nn.NLLLoss()
self.dis_criterion = nn.CrossEntropyLoss()
# DataLoader
self.gen_data = GenDataIter(self.gen.sample(cfg.batch_size, cfg.batch_size, self.dis))
self.dis_data = DisDataIter(self.gen_data.random_batch()['target'], self.train_data.random_batch()['target'])
# Metrics
self.bleu = BLEU(test_text=tensor_to_tokens(self.gen_data.target, self.index_word_dict),
real_text=tensor_to_tokens(self.test_data.target, self.test_data.index_word_dict), gram=3)
self.self_bleu = BLEU(test_text=tensor_to_tokens(self.gen_data.target, self.index_word_dict),
real_text=tensor_to_tokens(self.gen_data.target, self.index_word_dict),
gram=3)
def __init__(self, opt):
super(LeakGANInstructor, self).__init__(opt)
# generator, discriminator
self.gen = LeakGAN_G(cfg.gen_embed_dim, cfg.gen_hidden_dim,
cfg.vocab_size, cfg.max_seq_len, cfg.padding_idx,
cfg.goal_size, cfg.step_size, cfg.CUDA)
self.dis = LeakGAN_D(cfg.dis_embed_dim,
cfg.vocab_size,
cfg.padding_idx,
gpu=cfg.CUDA)
self.init_model()
# optimizer
mana_params, work_params = self.gen.split_params()
mana_opt = optim.Adam(mana_params, lr=cfg.gen_lr)
work_opt = optim.Adam(work_params, lr=cfg.gen_lr)
self.gen_opt = [mana_opt, work_opt]
self.dis_opt = optim.Adam(self.dis.parameters(), lr=cfg.dis_lr)
# Criterion
self.mle_criterion = nn.NLLLoss()
self.dis_criterion = nn.CrossEntropyLoss()
# DataLoader
self.gen_data = GenDataIter(
self.gen.sample(cfg.batch_size, cfg.batch_size, self.dis))
self.dis_data = DisDataIter(self.gen_data.random_batch()['target'],
self.oracle_data.random_batch()['target'])
self.dis_eval_data = DisDataIter(
self.gen_data.random_batch()['target'],
self.oracle_data.random_batch()['target'])
def __init__(self, opt):
super(LeakGANInstructor, self).__init__(opt)
# generator, discriminator
self.gen = LeakGAN_G(cfg.gen_embed_dim, cfg.gen_hidden_dim,
cfg.vocab_size, cfg.max_seq_len, cfg.padding_idx,
cfg.goal_size, cfg.step_size, cfg.CUDA)
self.dis = LeakGAN_D(cfg.dis_embed_dim,
cfg.vocab_size,
cfg.padding_idx,
gpu=cfg.CUDA)
self.init_model()
# optimizer
mana_params, work_params = self.gen.split_params()
mana_opt = optim.Adam(mana_params, lr=cfg.gen_lr)
work_opt = optim.Adam(work_params, lr=cfg.gen_lr)
self.gen_opt = [mana_opt, work_opt]
self.dis_opt = optim.Adam(self.dis.parameters(), lr=cfg.dis_lr)