pt
Building blocks and tools to construct, train, and manage PyTorch models.
Cast your data into tensors, embed your features, combine them (optionally extracting feature importance), and pass them through repeated residual layers of flexible architecture. Custom loss functions and re-parameterized distributions allow you to make probabilistic predictions. The training loop is conveniently abstracted away, allowing for (ready-to-use or custom-made) callbacks and checkpoints. Finally, trained models can be saved and (re)loaded.
Note
The tools in this subpackage are mostly geared towards tabular and sequence data. Image processing is not currently the main focus.