torchdata

torchdata is PyTorch oriented library focused on data processing and input pipelines in general.

It extends torch.utils.data.Dataset and equips it with functionalities known from tensorflow.data like map or cache (with some additions unavailable in aforementioned).

All of that with minimal interference (single call to super().__init__()) in original PyTorch's datasets.

Functionalities overview:

  • Use map, apply, reduce or filter
  • cache data in RAM/disk/your own method (even partially, say first 20%)
  • Full PyTorch's Dataset and IterableDataset support (including torchvision)
  • General torchdata.maps like Flatten or Select
  • Extensible interface (your own cache methods, cache modifiers, maps etc.)
  • Concrete torchdata.datasets designed for file reading and other general tasks

Quick examples

  • Create image dataset, convert it to Tensors, cache and concatenate with smoothed labels:
import torchdata
import torchvision

class Images(torchdata.Dataset): # Different inheritance
    def __init__(self, path: str):
        super().__init__() # This is the only change
        self.files = [file for file in pathlib.Path(path).glob("*")]

    def __getitem__(self, index):
        return Image.open(self.files[index])

    def __len__(self):
        return len(self.files)


images = Images("./data").map(torchvision.transforms.ToTensor()).cache()

You can concatenate above dataset with another (say labels) and iterate over them as per usual:

for data, label in images | labels:
    # Do whatever you want with your data
  • Cache first 1000 samples in memory, save the rest on disk in folder ./cache:
images = (
    ImageDataset.from_folder("./data").map(torchvision.transforms.ToTensor())
    # First 1000 samples in memory
    .cache(torchdata.modifiers.UpToIndex(1000, torchdata.cachers.Memory()))
    # Sample from 1000 to the end saved with Pickle on disk
    .cache(torchdata.modifiers.FromIndex(1000, torchdata.cachers.Pickle("./cache")))
    # You can define your own cachers, modifiers, see docs
)

To see what else you can do please check torchdata documentation

Installation

pip

Latest release:

pip install --user torchdata

Nightly:

pip install --user torchdata-nightly

Docker

CPU standalone and various versions of GPU enabled images are available
at dockerhub.

For CPU quickstart, issue:

docker pull szymonmaszke/torchdata:18.04

Nightly builds are also available, just prefix tag with nightly_. If you are going for GPU image make sure you have
nvidia/docker installed and it's runtime set.

GitHub