def load_dataloader(self) -> None:
"""Loads data loader for training and test."""
self.logger.info("Load dataset")
if self.config.model == "snp":
self.train_loader = torch.utils.data.DataLoader(
npr.SequentialGPDataset(train=True,
seq_len=20,
**self.config.train_dataset_params),
shuffle=True,
batch_size=16)
self.test_loader = torch.utils.data.DataLoader(
npr.SequentialGPDataset(train=False,
seq_len=20,
**self.config.test_dataset_params),
shuffle=False,
batch_size=1)
else:
self.train_loader = torch.utils.data.DataLoader(npr.GPDataset(
train=True, **self.config.train_dataset_params),
shuffle=True,
batch_size=16)
self.test_loader = torch.utils.data.DataLoader(npr.GPDataset(
train=False, **self.config.test_dataset_params),
shuffle=False,
batch_size=1)
self.logger.info(f"Train dataset size: {len(self.train_loader)}")
self.logger.info(f"Test dataset size: {len(self.test_loader)}")
def test_num_context(self):
dataset = npr.GPDataset(train=True, **self.params)
self.assertLessEqual(dataset.num_context,
self.params["num_context_max"])
dataset = npr.GPDataset(train=False, **self.params)
self.assertLessEqual(dataset.num_context,
self.params["num_context_max"])
def _small_case_target(self, train):
indices = [0, 1, 2]
dataset = npr.GPDataset(train=train, **self.params)
num_context_max = self.params["num_context_max"]
num_target_min = self.params["num_target_min"]
x_dim = self.params["x_dim"]
y_dim = self.params["y_dim"]
dataset.generate_dataset()
x_ctx, y_ctx, x_tgt, y_tgt = dataset[indices]
self.assertEqual(x_ctx.size(0), len(indices))
self.assertLessEqual(x_ctx.size(1), num_context_max)
self.assertEqual(x_ctx.size(2), x_dim)
self.assertEqual(y_ctx.size(0), len(indices))
self.assertLessEqual(y_ctx.size(1), num_context_max)
self.assertEqual(y_ctx.size(2), y_dim)
self.assertEqual(x_tgt.size(0), len(indices))
self.assertEqual(x_tgt.size(1), num_target_min)
self.assertEqual(x_tgt.size(2), x_dim)
self.assertEqual(y_tgt.size(0), len(indices))
self.assertEqual(y_tgt.size(1), num_target_min)
self.assertEqual(y_tgt.size(2), y_dim)
def test_generate_with_resample_params(self):
dataset = npr.GPDataset(train=True, **self.params)
dataset.generate_dataset(resample_params=True)
indices = [0, 1, 2]
x_ctx, y_ctx, x_tgt, y_tgt = dataset[indices]
num_context_max = self.params["num_context_max"]
num_target_max = self.params["num_target_max"]
x_dim = self.params["x_dim"]
y_dim = self.params["y_dim"]
self.assertEqual(x_ctx.size(0), len(indices))
self.assertLessEqual(x_ctx.size(1), num_context_max)
self.assertEqual(x_ctx.size(2), x_dim)
self.assertEqual(y_ctx.size(0), len(indices))
self.assertLessEqual(y_ctx.size(1), num_context_max)
self.assertEqual(y_ctx.size(2), y_dim)
self.assertEqual(x_tgt.size(0), len(indices))
self.assertLessEqual(x_tgt.size(1), num_target_max)
self.assertEqual(x_tgt.size(2), x_dim)
self.assertEqual(y_tgt.size(0), len(indices))
self.assertLessEqual(y_tgt.size(1), num_target_max)
self.assertEqual(y_tgt.size(2), y_dim)
def test_len(self):
dataset = npr.GPDataset(train=True, **self.params)
self.assertEqual(len(dataset), self.params["batch_size"])