def main2(input_file, output_file):
with open(input_file) as f_in:
data = json.load(f_in)
tmp = list()
tr = Translator()
for elem in data['data']:
trans = tr.translate(elem)
if not list(trans.keys()):
continue
key = list(trans.keys())[0]
val = trans[key][0]
tmp.append({'word': elem, 'translations': {key: val}})
data['data'] = tmp
with open(output_file, 'w+') as f_out:
f_out.write(json.dumps(data))
def post(self):
try:
data = json.loads(self.request.body)
except json.JSONDecodeError:
logging.warning(f"get incorrect body: {self.request.body}")
self.write(json.dumps({'error': 'incorrect-format'}))
return
if 'word' not in data:
logging.warning(f"get incorrect body: {self.request.body}")
self.write(json.dumps({'error': 'incorrect-format'}))
return
word = data['word'].lower().strip()
if not check_english_word(word):
logging.warning(f"get incorrect word: {word}")
self.write(json.dumps({'error': 'incorrect-format'}))
return
try:
translation = Translator().translate(word)
except RuntimeError:
logging.error(f"translation error")
self.write(json.dumps({'error': 'translation-error'}))
return
logging.debug(f"get translation {translation} for word {data['word']}")
self.write(
json.dumps({
'result': 'ok',
'data': {
'translations': translation
}
}))
def main4(input_file, output_file):
with open(input_file) as f_in:
data = json.load(f_in)
tmp = list()
tr = Translator()
for elem in data['data']:
if ' ' in elem['word']:
continue
key = list(elem['translations'].keys())[0]
val = elem['translations'][key][0]
tmp.append({
'word': elem['word'].lower(),
'translations': {
key: [val]
}
})
data['data'] = tmp
with open(output_file, 'w+') as f_out:
f_out.write(json.dumps(data))
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
parentdir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
grandfatherdir = os.path.dirname(
os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
sys.path.insert(0, parentdir)
sys.path.insert(0, grandfatherdir)
from utils.query_util import tokenize
from utils.translator import Translator
import dmn.char.dmn_data_utils as dmn_data_utils
from dmn.char.dmn_plus import Config, DMN_PLUS
translator = Translator()
EPOCH = 5
def prepare_data(args, config):
# train, valid, word_embedding, word2vec, updated_embedding, max_q_len, max_input_len, max_sen_len, \
# num_supporting_facts, vocab_size, candidate_size, candid2idx, \
# idx2candid, w2idx, idx2w = dmn_data_utils.load_data(
# config, split_sentences=True)
train_data, val_data, test_data, metadata = dmn_data_utils.load_data(
config, split_sentences=True)
# metadata = dict()
data = dict()
data['train'] = train_data
data['valid'] = val_data
def train(self):
""" Main training method for the Trainer class """
print("Starting training for {} epoch(s)".format(self.max_num_epochs -
self.epoch))
if not self.params.boost_warmup:
hard_training_instances = []
for epoch in range(self.max_num_epochs):
self.epoch = epoch
print("Epoch {}/{}".format(epoch + 1, self.max_num_epochs))
# train the model the train set
epoch_start_time = time.time()
# Make a copy of train_iter, add new examples to it (if boost==True),
# and pass it into train_epoch()
data_iterator = self.train_iter
# If boost==True and epochs are past warmup, perform boosting
if self.params.boost and epoch + 1 > self.params.boost_warmup:
print("Boosting....")
# make `Example` objects for all hard training instances
example_objs = self.create_example_objs(
hard_training_instances)
# Add the new hard training instances to the original training data
# thereby `boosting` the dataset with hard training examples
existing_data = self.train_iter.data()
existing_data.extend(example_objs)
# Create new Dataset and iterator on the boosted data
data_iterator = self.create_boosted_dataset(existing_data)
train_loss_avg, hard_training_instances = self.train_epoch(
data_iterator)
# write epoch statistics to Tensorboard
self.summary_writer.add_scalar("train/avg_loss_per_epoch",
train_loss_avg, self.epoch)
self.summary_writer.add_scalar("train/avg_perplexity_epoch",
math.exp(train_loss_avg),
self.epoch)
epoch_end_time = time.time()
epoch_mins, epoch_secs = self.epoch_time(epoch_start_time,
epoch_end_time)
print(
f'Epoch: {epoch+1:02} | Avg Train Loss: {train_loss_avg} | Perpelxity: {math.exp(train_loss_avg)} | Time: {epoch_mins}m {epoch_secs}s'
)
# validate the model on the dev set
val_start_time = time.time()
val_loss_avg = self.validate()
val_end_time = time.time()
val_mins, val_secs = self.epoch_time(val_start_time, val_end_time)
# write validation statistics to Tensorboard
self.summary_writer.add_scalar("val/loss", val_loss_avg,
self.epoch)
self.summary_writer.add_scalar("val/perplexity",
math.exp(val_loss_avg), self.epoch)
# TODO: write translations to Tensorboard
# every `decode_every_num_epochs` epochs, write out translations using Greedy Decoding
# to Tensorboard
if (self.epoch + 1) % self.decode_every_num_epochs == 0:
print("Performing Greedy Decoding...")
num_translations = 5
dev_iter = copy.copy(self.dev_iter)
decoder = Translator(
model=self.model,
dev_iter=list(dev_iter)[:num_translations],
params=self.params,
device=self.params.device)
translations = decoder.greedy_decode(max_len=100)
translations = [
" ".join(translation) for translation in translations
]
for translation in translations:
self.summary_writer.add_text("transformer/translation",
translation, self.epoch)
print(
f'Avg Val Loss: {val_loss_avg} | Val Perplexity: {math.exp(val_loss_avg)} | Time: {val_mins}m {val_secs}s'
)
print('\n')
# use a scheduler in order to decay learning rate hasn't improved
if self.scheduler is not None:
self.scheduler.step(val_loss_avg)
is_best = val_loss_avg < self.best_val_loss
optim_dict = self.optimizer._optimizer.state_dict() if isinstance(
self.optimizer,
ScheduledOptimizer) else self.optimizer.state_dict()
# save checkpoint
self.save_checkpoint(
{
"epoch": epoch + 1,
"state_dict": self.model.state_dict(),
"optim_dict": optim_dict
},
is_best=is_best,
checkpoint=self.params.model_dir + "/checkpoints/")
if is_best:
print("- Found new lowest loss!")
self.best_val_loss = val_loss_avg
replace_unk=False,
phrase_table='',
verbose=True,
dump_beam='',
n_best=1,
batch_type='sents',
gpu=0)
fields, model, model_opt = load_test_model(opt, args)
scorer = GNMTGlobalScorer.from_opt(opt)
out_file = codecs.open(opt.output, 'w+', 'utf-8')
translator = Translator.from_opt(model,
fields,
opt,
model_opt,
args,
global_scorer=scorer,
out_file=out_file,
report_align=opt.report_align,
report_score=False,
logger=None)
res = []
n = 1
with open(args.input_file, 'r') as f:
lines = f.readlines()
lines = [line.strip() for line in lines]
translated = translator.translate(lines, batch_size=args.batch_size)
for i in range(len(translated[1])):
res.append(translated[1][i][0])
if args.output_file:
def main(params, greedy, beam_size, test):
"""
The main function for decoding a trained MT model
Arguments:
params: parameters related to the `model` that is being decoded
greedy: whether or not to do greedy decoding
beam_size: size of beam if doing beam search
"""
print("Loading dataset...")
_, dev_iter, test_iterator, DE, EN = load_dataset(params.data_path,
params.train_batch_size,
params.dev_batch_size)
de_size, en_size = len(DE.vocab), len(EN.vocab)
print("[DE Vocab Size: ]: {}, [EN Vocab Size]: {}".format(
de_size, en_size))
params.src_vocab_size = de_size
params.tgt_vocab_size = en_size
params.sos_index = EN.vocab.stoi["<s>"]
params.pad_token = EN.vocab.stoi["<pad>"]
params.eos_index = EN.vocab.stoi["</s>"]
params.itos = EN.vocab.itos
device = torch.device('cuda' if params.cuda else 'cpu')
params.device = device
# make the Seq2Seq model
model = make_seq2seq_model(params)
# load the saved model for evaluation
if params.average > 1:
print("Averaging the last {} checkpoints".format(params.average))
checkpoint = {}
checkpoint["state_dict"] = average_checkpoints(params.model_dir,
params.average)
model = Trainer.load_checkpoint(model, checkpoint)
else:
model_path = os.path.join(params.model_dir + "checkpoints/",
params.model_file)
print("Restoring parameters from {}".format(model_path))
model = Trainer.load_checkpoint(model, model_path)
# evaluate on the test set
if test:
print("Doing Beam Search on the Test Set")
test_decoder = Translator(model, test_iterator, params, device)
test_beam_search_outputs = test_decoder.beam_decode(
beam_width=beam_size)
test_decoder.output_decoded_translations(
test_beam_search_outputs,
"beam_search_outputs_size_test={}.en".format(beam_size))
return
# instantiate a Translator object to translate SRC langauge to TRG language using Greedy/Beam Decoding
decoder = Translator(model, dev_iter, params, device)
if greedy:
print("Doing Greedy Decoding...")
greedy_outputs = decoder.greedy_decode(max_len=100)
decoder.output_decoded_translations(greedy_outputs,
"greedy_outputs.en")
print("Evaluating BLEU Score on Greedy Tranlsation...")
subprocess.call([
'./utils/eval.sh', params.model_dir + "outputs/greedy_outputs.en"
])
if beam_size:
print("Doing Beam Search...")
beam_search_outputs = decoder.beam_decode(beam_width=beam_size)
decoder.output_decoded_translations(
beam_search_outputs,
"beam_search_outputs_size={}.en".format(beam_size))
print("Evaluating BLEU Score on Beam Search Translation")
subprocess.call([
'./utils/eval.sh', params.model_dir +
"outputs/beam_search_outputs_size={}.en".format(beam_size)
])
