def valid(model, valid_dataset, step):
model.eval()
total_loss, total = 0.0, 0
hypothesis, references = [], []
for batch in valid_dataset:
loss = model(batch.src, batch.tgt).mean()
total_loss += loss.data
total += 1
predictions = parallel_beam_search(opt, model.module.model, batch, valid_dataset.fields)
hypothesis += [valid_dataset.fields["tgt"].decode(p) for p in predictions]
references += [valid_dataset.fields["tgt"].decode(t) for t in batch.tgt]
bleu = calculate_bleu(hypothesis, references)
logger.info("Valid loss: %.2f\tValid Beam BLEU: %3.2f" % (total_loss / total, bleu))
checkpoint = {"model": model.module.model.state_dict(), "opt": opt}
saver.save(checkpoint, printing_opt(opt), step, bleu, total_loss / total)
import torch.cuda
from beaver.data import build_dataset
from beaver.infer import beam_search
from beaver.loss import WarmAdam, LabelSmoothingLoss
from beaver.model import NMTModel
from beaver.utils import SaverTest
from beaver.utils import calculate_bleu
from beaver.utils import parseopt, get_device, printing_opt
from beaver.utils.metric import calculate_rouge
logging.basicConfig(format="%(asctime)s - %(message)s", level=logging.INFO)
opt = parseopt.parse_train_args()
device = get_device()
logging.info("\n" + printing_opt(opt))
saver = SaverTest(opt)
def write_file(hypothesis, references, step):
with open("hypothesis_" + str(step), "w", encoding="UTF-8") as out_file:
out_file.write("\n".join(hypothesis))
out_file.write("\n")
with open("references_" + str(step), "w", encoding="UTF-8") as out_file:
out_file.write("\n".join(references))
out_file.write("\n")
def valid(model, criterion_cn, criterion_en, valid_dataset, step):
from beaver.utils import Saver, Loader
from beaver.utils import calculate_bleu
from beaver.utils import parseopt, get_device, get_logger, printing_opt
parser = argparse.ArgumentParser()
parseopt.data_opts(parser)
parseopt.train_opts(parser)
parseopt.model_opts(parser)
opt = parser.parse_args()
device = get_device()
logger = get_logger()
logger.info("\n" + printing_opt(opt))
saver = Saver(save_path=opt.model_path, max_to_keep=opt.max_to_keep)
loader = Loader(opt.model_path, opt, logger)
def valid(model, valid_dataset, step):
model.eval()
total_loss, total = 0.0, 0
hypothesis, references = [], []
for batch in valid_dataset:
loss = model(batch.src, batch.tgt).mean()
total_loss += loss.data
total += 1