class Config(Optimizer.Config):
optimizer: Union[
SGD.Config,
Adam.Config,
AdamW.Config,
Adagrad.Config,
RAdam.Config,
Lamb.Config,
] = SGD.Config()
start: int = 10
frequency: int = 5
swa_learning_rate: Optional[float] = 0.05
def test_user_embedding_updates(self):
"""Verify the user embeddings learn independently."""
task_class = DocumentClassificationTask
pytext_config = self._get_pytext_config(
test_file_name=TestFileName.TEST_PERSONALIZATION_SINGLE_USER_TSV,
task_class=task_class,
model_class=PersonalizedDocModel,
)
# SGD changes only the user embeddings which have non-zero gradients.
pytext_config.task.trainer.optimizer = SGD.Config()
p13n_task = task_class.from_config(pytext_config.task)
orig_user_embedding_weights = copy.deepcopy(
p13n_task.model.user_embedding.weight)
p13n_model, _ = p13n_task.train(pytext_config)
trained_user_embedding_weights = p13n_model.user_embedding.weight
self.assertEqual(
len(orig_user_embedding_weights),
2,
"There should be 2 user embeddings, including the unknown user.",
)
self.assertEqual(
len(orig_user_embedding_weights),
len(trained_user_embedding_weights),
"Length of user embeddings should not be changed by the training.",
)
# Verify that the training changes only 1 user embedding in the p13n_model.
self.assertTrue(
torch.equal(orig_user_embedding_weights[0],
trained_user_embedding_weights[0]),
"Unknown user embedding should not change.",
)
self.assertFalse(
torch.equal(orig_user_embedding_weights[1],
trained_user_embedding_weights[1]),
"The only user embedding should change.",
)
class Config(Optimizer.Config):
optimizer: Union[SGD.Config, Adam.Config] = SGD.Config()
start: int = 10
frequency: int = 5
swa_learning_rate: float = 0.05