mila-iqia
Save best model in machine translation
I find the best model saved by the machine translation example can't be load. I think we need to change sampling.py\BleuValidator()\_save_model()
numpy.savez(
model.path, **self.main_loop.model.get_parameter_dict())
to
numpy.savez(
model.path, **self.main_loop.model.get_parameter_values())
According to the blocks document on model: get_parameter_dict() returns a dictionary as dictionary{hierarchical name: shared variable} while get_parameter_values() returns a dictionary as dictionary{hierarchical name: numpy.ndarray}, where later one should be better option.
In fact, the custom serialization method used in this demo should be dropped. Since about two months Blocks has a new implementation of serialization, that should meet all the requirements for MT.
On 25 April 2016 at 17:27, Qiang Zhang [email protected] wrote:
I find the best model saved by the machine translation example can't be load. I think we need to change
numpy.savez( model.path, **self.main_loop.model.get_parameter_dict())to
numpy.savez( model.path, **self.main_loop.model.get_parameter_values())According to the blocks document on model: get_parameter_dict() returns a dictionary as dictionary{hierarchical name: shared variable} while get_parameter_values() returns a dictionary as dictionary{hierarchical name: numpy.ndarray}, where later one should be better option.
— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/mila-udem/blocks-examples/issues/97
A temporary fix for this could be:
def load_model_from_bleu_file(self, model, bleu_file_path):
assert os.path.exists(bleu_file_path)
with closing(numpy.load(bleu_file_path)) as source:
param_values = {}
print source.items()
for name, parameter in source.items():
if name != 'pkl':
name_ = name.replace(BRICK_DELIMITER, '/')
if not name_.startswith('/'):
name_ = '/' + name_
param_values[name_] = parameter.flatten()[0].get_value()
model.set_parameter_values(param_values)
