Number of data points of DataLoader with SubsetRandomSampler

1

We have a DataLoader that is using SubsetRandomSampler. Is it possible to tell the number of datapoints in the loader (which are sampled) from the DataLoader?

2

You can print the len of the internal .sampler as seen here:

N = 100
dataset = torch.utils.data.TensorDataset(torch.randn(N, 1))
sampler = torch.utils.data.sampler.SubsetRandomSampler(indices=torch.arange(N//2))
loader = torch.utils.data.DataLoader(dataset, sampler=sampler, batch_size=7)

print(len(loader))
# 8
# 7 full batches and 1 batch with a single element: 7*7+1*1 = 50

print(len(loader.sampler))
# 50 number of samples
1 Like
3

Perfect, it seems that it’s even working when there is no sampler and return the value of the whole dataset.

You’re the best, @ptrblck