def wrap_dataset(self, dset, savedir, args):
# save the data
save_bball_data(dset, savedir, args.override_data)
dset = load_bball_data(savedir, args.target)
#Deal with uneven lengths breaking DataLoader
def pad_data(x):
max_len = max([len(sig[0]) for sig in x])
new_x = []
for data in x:
new_data = []
for datum in data:
datum = np.append(datum, np.zeros(max_len - len(datum)))
new_data.append([float(i) for i in datum])
new_x.append(new_data)
return new_x
def my_collate(batch):
data = [item[0] for item in batch]
lengths = [len(d[0]) for d in data]
data = torch.stack(
[torch.Tensor(x[0]).float() for x in pad_data(data)])
target = [item[1] for item in batch]
target = torch.FloatTensor(target)
lens = torch.LongTensor(lengths)
return [data, target, lens]
return DataLoader(dset,
batch_size=args.batch_size,
collate_fn=my_collate,
batch_first=True,
shuffle=True,
num_workers=args.num_workers)
def wrap_dataset(self, dset, savedir, args):
# save the data
save_bball_data(dset, savedir, args.override_data)
# return dataloader of the saved data
dset = load_bball_data(savedir, args.target)
return DataLoader(dset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.num_workers)