Different strategies to implement noiseless regression in Python. The problem is a 3-D Euclidean grid, split into different case sizes to select from if you can't run big cases:
- full: 141 × 66 × 5.
- mid: 36 × 66 × 5
- small: 36 × 33 × 5
- tiny: 18 × 16 × 5
The regression options are:
- Gaussian Processes: (all with ARD kernels, apparently this problem won't fit otherwise)
- scikit
- gpflow (based on tensorflow)
- GPYTorch (based on PyTorch)
- Neural networks
- classical fully connected (keras)
- multi-head self-attention (keras)