def test_compose(self):
wav = np.array([1, 3])
assert torch.allclose(
transforms.Compose([
transforms.ZeroMean(),
transforms.ToTensor(),
])(wav), torch.tensor([[-1, 1]], dtype=torch.float))
from torch.nn import CrossEntropyLoss
from torch.utils.data import DataLoader
from torch.utils.data import random_split
import yews.datasets as dsets
import yews.transforms as transforms
from yews.models import Cpic
from yews.train import Trainer
if __name__ == '__main__':
# Preprocessing
waveform_transform = transforms.Compose([
transforms.ZeroMean(),
transforms.SoftClip(1e-4),
transforms.ToTensor(),
])
# Prepare dataset
dset = dsets.Wenchuan(path='.', download=True,
sample_transform=waveform_transform)
# Split datasets into training and validation
train_length = int(len(dset) * 0.8)
val_length = len(dset) - train_length
train_set, val_set = random_split(dset, [train_length, val_length])
# Prepare dataloaders
train_loader = DataLoader(train_set, batch_size=100, shuffle=True, num_workers=4)
val_loader = DataLoader(val_set, batch_size=1000, shuffle=False, num_workers=8)
def test_to_tensor_shape(self):
wav = np.array([1])
assert transforms.ToTensor()(wav).shape == (1,1)
def test_to_tensor_dtype(self):
wav = np.array([1])
assert torch.allclose(torch.tensor([[1]], dtype=torch.float),
transforms.ToTensor()(wav))
