def test_decoding_2(self):
src = torch.rand(20, 3)
trg = torch.rand(354, 3)
src1 = torch.rand(20, 4)
trg1 = torch.rand(354, 4)
d = decoding_function(self.informer, src, trg, 5, src1, trg1, 1, 20,
336, "cpu")
self.assertEqual(d.shape[0], 1)
self.assertEqual(d.shape[1], 336)
def generate_decoded_predictions(
model: Type[TimeSeriesModel],
test_data: CSVTestLoader,
forecast_start_idx: int,
device: torch.device,
history_dim: torch.Tensor,
hours_to_forecast: int,
decoder_params: Dict,
multi_targets: int = 1,
targs: Union[bool, torch.Tensor] = False) -> torch.Tensor:
probabilistic = False
scaler = None
if test_data.no_scale:
scaler = test_data
if decoder_params is not None:
if "probabilistic" in decoder_params:
probabilistic = True
real_target_tensor = (torch.from_numpy(
test_data.df[forecast_start_idx:].to_numpy()).to(device).unsqueeze(
0).to(model.device))
if targs:
src = history_dim
src0 = src[0]
trg = targs
decoder_seq_len = model.params["model_params"]["label_len"]
end_tensor = decoding_function(
model.model, src0, trg[1],
model.params["dataset_params"]["forecast_length"], src[1],
trg[0], 1, decoder_seq_len, hours_to_forecast, device)
else:
end_tensor = decoding_functions[
decoder_params["decoder_function"]](
model.model,
history_dim,
hours_to_forecast,
real_target_tensor,
decoder_params["unsqueeze_dim"],
output_len=model.params["dataset_params"]
["forecast_length"],
multi_targets=multi_targets,
device=model.device,
probabilistic=probabilistic,
scaler=scaler)
if probabilistic:
end_tensor_mean = end_tensor[0][:, :,
0].view(-1).to("cpu").detach()
return end_tensor_mean, end_tensor[1]
elif isinstance(end_tensor, tuple):
e = end_tensor[0][:, :, 0].view(-1).to(
"cpu").detach(), end_tensor[1][:, :,
0].view(-1).to("cpu").detach()
return e
if multi_targets == 1:
end_tensor = end_tensor[:, :, 0].view(-1)
return end_tensor.to("cpu").detach()
def test_decoding_3(self):
informer_model2 = Informer(3, 3, 3, 48, 24, 12, factor=1)
src = torch.rand(1, 48, 3)
trg = torch.rand(1, 362, 3)
src1 = torch.rand(1, 48, 4)
trg1 = torch.rand(1, 362, 4)
d = decoding_function(informer_model2, src, trg, 12, src1, trg1, 1, 36,
336, "cpu")
self.assertEqual(d.shape[0], 1)
self.assertEqual(d.shape[1], 336)
def test_decoding_2(self):
src = torch.rand(20, 3)
trg = torch.rand(355, 3)
src1 = torch.rand(20, 4)
trg1 = torch.rand(355, 4)
d = decoding_function(self.informer, src, trg, 5, src1, trg1, 1, 20,
336, "cpu")
self.assertEqual(d.shape[0], 1)
self.assertEqual(d.shape[1], 336)
self.assertNotEqual(d[0, 0, 0].item(), d[0, 1, 0].item())
self.assertNotAlmostEqual(d[0, 0, 0].item(), d[0, 330, 0].item())
self.assertNotAlmostEqual(d[0, 20, 0].item(), d[0, 333, 0].item())
self.assertNotAlmostEqual(d[0, 300, 0].item(), d[0, 334, 0].item())
self.assertNotAlmostEqual(d[0, 20, 0].item(), trg[20, 0].item())
self.assertNotAlmostEqual(d[0, 21, 0].item(), trg[21, 0].item())
