PyTorch DataLoader demo. Trains a neural network to fit spiral data.
This notebook is a demonstration of workflow for big-data handling in PyTorch.
The PyTorch DataLoader class is a mechanism to serve batches of data to a machine learning agent. It can be arranged to work with data already in memory but also to serve data from disk on-the-fly.
When a sizeable data set will not fit into RAM the DataLoader class offers a solution by drawing manageable batches from disk. This is accomplished in parallel by multiple workers thereby keeping a demanding GPU supplied with mini-batches of data.
What follows is a small-data exercise using big-data techniques.