"Dirty-MNIST from \"Deterministic Neural Networks with Appropriate Inductive Biases Capture Epistemic and Aleatoric Uncertainty\""
Project description
DDU's Dirty-MNIST
You'll never want to use MNIST again for OOD or AL.
Install
pip install ddu_dirty_mnist
How to use
After installing, you get a Dirty-MNIST train or test set just like you would for MNIST in PyTorch.
# gpu
import ddu_dirty_mnist
dirty_mnist_train = ddu_dirty_mnist.DirtyMNIST(".", train=True, download=True, device="cuda")
dirty_mnist_test = ddu_dirty_mnist.DirtyMNIST(".", train=False, download=True, device="cuda")
len(dirty_mnist_train), len(dirty_mnist_test)
(120000, 30000)
Here is how to create torch.utils.data.DataLoader
, see the documentation for details.
# gpu
import torch
dirty_mnist_train_dataloader = torch.utils.data.DataLoader(
dirty_mnist_train,
batch_size=128,
shuffle=True,
num_workers=0,
pin_memory=False,
)
dirty_mnist_test_dataloader = torch.utils.data.DataLoader(
dirty_mnist_test,
batch_size=128,
shuffle=False,
num_workers=0,
pin_memory=False,
)
If you only care about Ambiguous-MNIST, you can use:
# gpu
import ddu_dirty_mnist
ambiguous_mnist_train = ddu_dirty_mnist.AmbiguousMNIST(".", train=True, download=True, device="cuda")
ambiguous_mnist_test = ddu_dirty_mnist.AmbiguousMNIST(".", train=False, download=True, device="cuda")
ambiguous_mnist_train, ambiguous_mnist_test
(Dataset AmbiguousMNIST
Number of datapoints: 60000
Root location: .,
Dataset AmbiguousMNIST
Number of datapoints: 20000
Root location: .)
Here is how to create torch.utils.data.DataLoader
, see the documentation for details.
# gpu
import torch
ambiguous_mnist_train_dataloader = torch.utils.data.DataLoader(
ambiguous_mnist_train,
batch_size=128,
shuffle=True,
num_workers=0,
pin_memory=False,
)
ambiguous_mnist_test_dataloader = torch.utils.data.DataLoader(
ambiguous_mnist_test,
batch_size=128,
shuffle=False,
num_workers=0,
pin_memory=False,
)
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
ddu_dirty_mnist-0.6.tar.gz
(11.9 kB
view hashes)
Built Distribution
Close
Hashes for ddu_dirty_mnist-0.6-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 41df526c21f089a2b4429f50177a1c50c026514ee2776e51dea8d45bff36d434 |
|
MD5 | b7a1b9808ac9b4657f79fc792f08730c |
|
BLAKE2b-256 | 8a852991824c7d77d57549e8f77ee003f337a48f62cb3de49b12d8f08d1a29a5 |