def __call__(self):
logger_util.add_model_indicators(self.model)
logger.add_indicator(Queue("train_loss", 20, True))
logger.add_indicator(Histogram("test_loss", True))
logger.add_indicator(Histogram("accuracy", True))
for _ in self.loop:
self._train()
self._test()
self.__log_model_params()
def __call__(self):
# Training and testing
logger_util.add_model_indicators(self.model)
logger.add_indicator(Queue("train_loss", 20, True))
logger.add_indicator(Histogram("test_loss", True))
logger.add_indicator(Histogram("accuracy", True))
logger.add_indicator(IndexedScalar('test_sample_loss'))
logger.add_indicator(IndexedScalar('test_sample_pred'))
test_data = np.array([d[0].numpy() for d in self.test_loader.dataset])
logger.save_numpy("test_data", test_data)
for _ in self.loop:
self._train()
self._test()
self.__log_model_params()
def add_model_indicators(model: torch.nn.Module, model_name: str = "model"):
for name, param in model.named_parameters():
if param.requires_grad:
logger.add_indicator(Histogram(f"{model_name}.{name}"))
logger.add_indicator(Histogram(f"{model_name}.{name}.grad"))
def startup(self):
logger_util.add_model_indicators(self.model)
logger.add_indicator(Queue("train_loss", 20, True))
logger.add_indicator(Histogram("test_loss", True))
logger.add_indicator(Histogram("accuracy", True))