How to Build a Streaming DataLoader with PyTorch

PyTorch Datasets are objects that have a single job: to return a single datapoint on request. The exact form of the datapoint varies between tasks: it could be a single image, a slice of a time series, a tabular record or something else entirely. These are then passed on to a Dataloader which handles batching of datapoints and parallelism. Before PyTorch 1.2 the only available dataset class was the original "map-style" dataset. This simply requires the user to inherit from the torch.utils.data.Dataset class and implement the __len__ and __getitem__ methods, where __getitem__ receives an index which is mapped to some item in your dataset.