def train(models, train_set, eval_set=None, silent=False):
"""Train all model for production and save them
Args:
models (list of str): Model names. Pass if you want to train a just a
set particular models
train_set (dg.enums.Dataset): Dataset to train on
eval_set (dg.enums.Dataset): Dataset to use for evaluation during
training.
silent (bool): Don't print details to standard out.
"""
config = Config()
model_dir = config.get_model_dir()
if not silent:
print('Model dir: ', model_dir)
bar(silent=silent)
for model_id in models:
model = config.models[model_id].set_params(
**config.get_params(model_id))
datasets = config.get_datasets(model.id)
train_set = (datasets[train_set.value] if isinstance(
train_set, Dataset) else train_set)
eval_set = (datasets[eval_set.value]
if isinstance(eval_set, Dataset) else eval_set)
train_model(model,
train_set=train_set,
eval_set=eval_set,
model_dir=model_dir,
save=True,
silent=silent)
bar(silent=silent)
def load(cls, model_dir):
"""Load the production model
Args:
model_dir (str): Path to the model dir from where we should load
the model.
"""
config = Config()
with io.open(os.path.join(model_dir, 'params.yaml')) as f:
params = yaml.safe_load(f)
model = cls(**params)
model.model_dir = config.get_model_dir(tensorflow=True)
model.estimator = model._create_estimator(model_dir)
return model