mantra.transforms
Transforms module
A set of base transforms for the MANTRA dataset. We make use of such transformations in our paper to enable the training on different neural-network architectures.
- class mantra.transforms.NodeIndex
This transform ensures the compatibility with
pytorch-geometricby changing the node/vertex indices to be zero-indexed.
- class mantra.transforms.RandomNodeFeatures(dimension=8)
Adds random node features to the dataset. The main purpose behind this transformation is to ensure compatibility with architectures that require node features, while also showing their respective shortcomings. In our dataset, unlike many others, node coordinates and the triangulations themselves are fully decoupled.
- __init__(dimension=8)
- class mantra.transforms.DegreeTransform