def get_w2v_model(fetch_from_s3=False,
corpus_name=TRAIN_CORPUS_NAME,
voc_size=VOCABULARY_MAX_SIZE,
vec_size=WORD_EMBEDDING_DIMENSION,
window_size=W2V_WINDOW_SIZE,
skip_gram=USE_SKIP_GRAM):
try:
model_resolver_factory = \
S3FileResolver.init_resolver(bucket_name=S3_MODELS_BUCKET_NAME, remote_dir=S3_W2V_REMOTE_DIR) \
if fetch_from_s3 else None
w2v_model = _get_w2v_model(
corpus_name=corpus_name,
voc_size=voc_size,
model_resolver_factory=model_resolver_factory,
vec_size=vec_size,
window_size=window_size,
skip_gram=skip_gram)
except ModelLoaderException:
raise ModelLoaderException('Word2vec model is absent. Please run `tools/train_w2v.py` to get the model.'
' WARNING: If you compare different dialog models, be sure that they'
' use the same w2v model (since each run of the w2v-trainer even with the same'
' parameters leads to different w2v models)')
return w2v_model
def train(init_path=None, is_reverse_model=False):
processed_train_corpus_path = get_processed_corpus_path(TRAIN_CORPUS_NAME)
processed_val_corpus_path = get_processed_corpus_path(
CONTEXT_SENSITIVE_VAL_CORPUS_NAME)
index_to_token_path = get_index_to_token_path(BASE_CORPUS_NAME)
index_to_condition_path = get_index_to_condition_path(BASE_CORPUS_NAME)
# check the existence of all necessary files before compiling the model
_look_for_saved_files(files_paths=[
processed_train_corpus_path, processed_val_corpus_path,
index_to_token_path
])
index_to_token = load_index_to_item(index_to_token_path)
index_to_condition = load_index_to_item(index_to_condition_path)
w2v_matrix = _get_w2v_embedding_matrix_by_corpus_path(
processed_train_corpus_path, index_to_token)
# get nn_model and train it
nn_model_resolver_factory = S3FileResolver.init_resolver(
bucket_name=S3_MODELS_BUCKET_NAME, remote_dir=S3_NN_MODEL_REMOTE_DIR)
nn_model, _ = get_nn_model(index_to_token,
index_to_condition,
model_init_path=init_path,
w2v_matrix=w2v_matrix,
resolver_factory=nn_model_resolver_factory,
is_reverse_model=is_reverse_model)
train_model(nn_model)
def get_w2v_embedding_matrix(tokenized_dialog_lines, index_to_token, add_start_end=False):
if add_start_end:
tokenized_dialog_lines = (
[SPECIAL_TOKENS.START_TOKEN] + line + [SPECIAL_TOKENS.EOS_TOKEN] for line in tokenized_dialog_lines)
w2v_resolver_factory = S3FileResolver.init_resolver(bucket_name=S3_MODELS_BUCKET_NAME, remote_dir=S3_W2V_REMOTE_DIR)
w2v_model = get_w2v_model(
TRAIN_CORPUS_NAME,
len(index_to_token),
model_resolver_factory=w2v_resolver_factory,
tokenized_lines=tokenized_dialog_lines,
vec_size=WORD_EMBEDDING_DIMENSION,
window_size=W2V_WINDOW_SIZE,
skip_gram=USE_SKIP_GRAM)
w2v_matrix = transform_w2v_model_to_matrix(w2v_model, index_to_token)
return w2v_matrix
def get_trained_model(reverse=False, fetch_from_s3=True):
if fetch_from_s3:
resolver_factory = S3FileResolver.init_resolver(
bucket_name=S3_MODELS_BUCKET_NAME, remote_dir=S3_NN_MODEL_REMOTE_DIR)
else:
resolver_factory = None
nn_model, model_exists = get_nn_model(
_get_index_to_token(fetch_from_s3),
_get_index_to_condition(fetch_from_s3),
resolver_factory=resolver_factory,
is_reverse_model=reverse)
if not model_exists:
raise Exception('Can\'t get the model. '
'Run tools/download_model.py first to get all required files or train it by yourself.')
return nn_model
def get_trained_model(reverse=False, fetch_from_s3=True):
if fetch_from_s3:
resolver_factory = S3FileResolver.init_resolver(
bucket_name=S3_MODELS_BUCKET_NAME,
remote_dir=S3_NN_MODEL_REMOTE_DIR)
else:
resolver_factory = None
nn_model, model_exists = get_nn_model(
index_to_token=_get_index_to_token(fetch_from_s3),
index_to_condition=_get_index_to_condition(fetch_from_s3),
resolver_factory=resolver_factory,
is_reverse_model=reverse)
if not model_exists:
raise FileNotFoundException(
'Can\'t get the pre-trained model. Run tools/download_model.py first '
'to get all required files or train it by yourself.')
return nn_model
def train(model_init_path=None,
is_reverse_model=False,
train_subset_size=None,
use_pretrained_w2v=USE_PRETRAINED_W2V_EMBEDDINGS_LAYER,
train_corpus_name=TRAIN_CORPUS_NAME,
context_sensitive_val_corpus_name=CONTEXT_SENSITIVE_VAL_CORPUS_NAME,
base_corpus_name=BASE_CORPUS_NAME,
s3_models_bucket_name=S3_MODELS_BUCKET_NAME,
s3_nn_model_remote_dir=S3_NN_MODEL_REMOTE_DIR,
prediction_mode_for_tests=PREDICTION_MODE_FOR_TESTS):
processed_train_corpus_path = get_processed_corpus_path(train_corpus_name)
processed_val_corpus_path = get_processed_corpus_path(
context_sensitive_val_corpus_name)
index_to_token_path = get_index_to_token_path(base_corpus_name)
index_to_condition_path = get_index_to_condition_path(base_corpus_name)
# check the existence of all necessary files before compiling the model
_look_for_saved_files(files_paths=[
processed_train_corpus_path, processed_val_corpus_path,
index_to_token_path
])
# load essentials for building model and training
index_to_token = load_index_to_item(index_to_token_path)
index_to_condition = load_index_to_item(index_to_condition_path)
token_to_index = {v: k for k, v in index_to_token.items()}
condition_to_index = {v: k for k, v in index_to_condition.items()}
training_data_param = ModelParam(value=get_training_dataset(
train_corpus_name, token_to_index, condition_to_index,
is_reverse_model, train_subset_size),
id=train_corpus_name)
val_sets_names = get_validation_sets_names()
validation_data_param = ModelParam(
value=get_validation_dataset_name_to_data(val_sets_names,
token_to_index,
condition_to_index,
is_reverse_model),
id=get_validation_data_id(val_sets_names))
w2v_model_param = ModelParam(value=get_w2v_model(), id=get_w2v_model_id()) if use_pretrained_w2v \
else ModelParam(value=None, id=None)
model_resolver_factory = S3FileResolver.init_resolver(
bucket_name=s3_models_bucket_name, remote_dir=s3_nn_model_remote_dir)
reverse_model = get_reverse_model(
prediction_mode_for_tests) if not is_reverse_model else None
# build CakeChatModel
cakechat_model = CakeChatModel(index_to_token,
index_to_condition,
training_data_param=training_data_param,
validation_data_param=validation_data_param,
w2v_model_param=w2v_model_param,
model_init_path=model_init_path,
model_resolver=model_resolver_factory,
is_reverse_model=is_reverse_model,
reverse_model=reverse_model,
horovod=hvd)
# train model
cakechat_model.train_model()