def set_eval_inference_latency_mode(self):
"""
Evaluate Inference Latency Mode
- Pipeline
1. read raw_data (DataReader)
2. load vocabs from checkpoint (DataReader, Token)
3. define raw_to_tensor_fn (DataReader, Token)
4. define and load model
5. run!
"""
data_reader, token_makers = self._create_data_and_token_makers()
# Token & Vocab
vocabs = utils.load_vocabs(self.model_checkpoint)
for token_name, token_maker in token_makers.items():
token_maker.set_vocab(vocabs[token_name])
text_handler = TextHandler(token_makers, lazy_indexing=False)
_, helpers = data_reader.read()
raw_examples = helpers["valid"]["examples"]
cuda_device = self.config.cuda_devices[
0] if self.config.use_gpu else None
raw_to_tensor_fn = text_handler.raw_to_tensor_fn(
data_reader, cuda_device=cuda_device)
# Model
model = self._create_model(token_makers,
checkpoint=self.model_checkpoint)
self.set_trainer(model)
return raw_examples, raw_to_tensor_fn
def set_predict_mode(self, preload=False):
"""
Predict Mode
- Pipeline
1. read raw_data (Argument)
2. load vocabs from checkpoint (DataReader, Token)
3. define raw_to_tensor_fn (DataReader, Token)
4. define and load model
5. run!
"""
data_reader, token_makers = self._create_data_and_token_makers()
# Token & Vocab
vocabs = utils.load_vocabs(self.model_checkpoint)
for token_name, token_maker in token_makers.items():
token_maker.set_vocab(vocabs[token_name])
text_handler = TextHandler(token_makers, lazy_indexing=False)
# Set predict config
if self.argument.interactive:
raw_features = {
feature_name: ""
for feature_name in data_reader.text_columns
}
else:
raw_features = {}
for feature_name in data_reader.text_columns:
feature = getattr(self.argument, feature_name, None)
# if feature is None:
# raise ValueError(f"--{feature_name} argument is required!")
raw_features[feature_name] = feature
cuda_device = self.config.cuda_devices[
0] if self.config.use_gpu else None
raw_to_tensor_fn = text_handler.raw_to_tensor_fn(
data_reader,
cuda_device=cuda_device,
helper=self.model_checkpoint.get("predict_helper", {}))
# Model
model = self._create_model(token_makers,
checkpoint=self.model_checkpoint)
self.set_trainer(model)
arguments = vars(self.argument)
if preload:
self.predict_settings = {
"raw_to_tensor_fn": raw_to_tensor_fn,
"arguments": arguments
}
else:
return raw_features, raw_to_tensor_fn, arguments
def set_eval_mode(self):
"""
Evaluate Mode
- Pipeline
1. read raw_data (DataReader)
2. load vocabs from checkpoint (DataReader, Token)
3. indexing tokens (DataReader, Token)
4. convert to DataSet (DataReader)
5. create DataLoader (DataLoader)
6. define and load model
7. run!
"""
data_reader, token_makers = self._create_data_and_token_makers()
# DataReader
datas, helpers = data_reader.read()
# Token & Vocab
vocabs = utils.load_vocabs(self.model_checkpoint)
for token_name, token_maker in token_makers.items():
token_maker.set_vocab(vocabs[token_name])
text_handler = TextHandler(token_makers, lazy_indexing=False)
text_handler.index(datas, data_reader.text_columns)
# iterator
vocab = vocabs[next(iter(vocabs))]
datasets = data_reader.convert_to_dataset(datas,
vocab,
helpers=helpers) # with name
self.config.iterator.cuda_devices = self.config.cuda_devices
_, valid_loader, _ = self._create_by_factory(
DataLoaderFactory,
self.config.iterator,
param={"datasets": datasets})
# Model
model = self._create_model(token_makers,
checkpoint=self.model_checkpoint)
self.set_trainer(model)
return valid_loader
