def main():
rnd_seed = None
if rnd_seed:
torch.manual_seed(rnd_seed)
np.random.seed(rnd_seed)
result_path = "../result_ccg_lrn_0p001_atten/"
dict_file = "../dataset/CCGbank/dict_word"
entity_file = "../dataset/CCGbank/dict_tag"
index2word = get_index2word(dict_file)
index2label = get_index2label(entity_file)
vocab_size = len(index2word)
label_size = len(index2label)
val_X, val_Y = minibatch_of_one_de('val')
test_X, test_Y = minibatch_of_one_de('test')
# Using word2vec pre-trained embedding
word_embedding_dim = 300
hidden_dim = 512
label_embedding_dim = 512
max_epoch = 30
# 0.001 is a good value
learning_rate = 0.001
attention = "fixed"
#attention = None
pretrained = None
gpu = True
if gpu and rnd_seed:
torch.cuda.manual_seed(rnd_seed)
gpu_no = 2
cuda_dev = torch.device("cuda:" + str(gpu_no))
##################
os.environ['OMP_NUM_THREADS'] = '1'
eval_output_file = open(os.path.join(result_path, "eval_beam_2_ckpt_11.txt"), "w+")
for epoch in range(11, 12):
load_model_filename = os.path.join(result_path, "ckpt_" + str(epoch) + ".pth")
batch_size = 1
machine = ner(word_embedding_dim, hidden_dim, label_embedding_dim, vocab_size, label_size, learning_rate=learning_rate, minibatch_size=batch_size, max_epoch=max_epoch, train_X=None, train_Y=None, val_X=val_X, val_Y=val_Y, test_X=test_X, test_Y=test_Y, attention=attention, gpu=gpu, gpu_no=gpu_no, pretrained=pretrained, load_model_filename=load_model_filename)
if gpu:
machine = machine.cuda(cuda_dev)
decode_method = "beam"
beam_size = 2
max_beam_size = label_size
accum_logP_ratio_low = 0.1
logP_ratio_low = 0.1
#agent = det_agent(max_beam_size, accum_logP_ratio_low, logP_ratio_low)
agent = None
# For German dataset, f_score_index_begin = 5 (because O_INDEX = 4)
# For toy dataset, f_score_index_begin = 4 (because {0: '<s>', 1: '<e>', 2: '<p>', 3: '<u>', ...})
# For CCG dataset, f_score_index_begin = 2 (because {0: _PAD, 1: _SOS, ...})
f_score_index_begin = 2
reward_coef_fscore = 1
reward_coef_beam_size = 0.02
# We don't evaluate on training set simply because it is too slow since we can't use mini-batch in adaptive beam search
val_fscore, val_beam_number, val_avg_beam_size = machine.evaluate(val_X, val_Y, index2word, index2label, "val", None, decode_method, beam_size, max_beam_size, agent, reward_coef_fscore, reward_coef_beam_size, f_score_index_begin, generate_episode=False, episode_save_path=None)
time_begin = time.time()
test_fscore, test_beam_number, test_avg_beam_size = machine.evaluate(test_X, test_Y, index2word, index2label, "test", None, decode_method, beam_size, max_beam_size, agent, reward_coef_fscore, reward_coef_beam_size, f_score_index_begin, generate_episode=False, episode_save_path=None)
time_end = time.time()
print_msg = "epoch %d, val F = %.6f, test F = %.6f, test time = %.6f" % (epoch, val_fscore, test_fscore, time_end - time_begin)
log_msg = "%d\t%f\t%f\t%d\t%d\t%f\t%f\t%f" % (epoch, val_fscore, test_fscore, val_beam_number, test_beam_number, val_avg_beam_size, test_avg_beam_size, time_end - time_begin)
print(print_msg)
print(log_msg, file=eval_output_file, flush=True)
# End for epoch
eval_output_file.close()
def main():
rnd_seed = None
if rnd_seed:
torch.manual_seed(rnd_seed)
np.random.seed(rnd_seed)
#data_path = "../dataset/German/"
result_path = "../result_ccg_lrn_0p001/"
if not os.path.exists(result_path):
os.makedirs(result_path)
batch_size = 32
dict_file = "../dataset/CCGbank/dict_word"
entity_file = "../dataset/CCGbank/dict_tag"
index2word = get_index2word(dict_file)
index2label = get_index2label(entity_file)
vocab_size = len(index2word)
label_size = len(index2label)
#print("label_size=",label_size)
train_X, train_Y = minibatch_de('train', batch_size)
# Using word2vec pre-trained embedding
word_embedding_dim = 300
hidden_dim = 512
label_embedding_dim = 512
max_epoch = 100
# 0.001 is a good value
learning_rate = 0.001
#attention = "fixed"
attention = None
pretrained = None
if pretrained == 'de64':
word_embedding_dim = 64
gpu = True
if gpu and rnd_seed:
torch.cuda.manual_seed(rnd_seed)
load_model_filename = None
machine = ner(word_embedding_dim,
hidden_dim,
label_embedding_dim,
vocab_size,
label_size,
learning_rate=learning_rate,
minibatch_size=batch_size,
max_epoch=max_epoch,
train_X=train_X,
train_Y=train_Y,
val_X=None,
val_Y=None,
test_X=None,
test_Y=None,
attention=attention,
gpu=gpu,
pretrained=pretrained,
load_model_filename=load_model_filename)
if gpu:
machine = machine.cuda()
shuffle = True
# Pure training, no evaluation
train_loss_list = machine.train(shuffle, result_path, False, None)