**This release of HexagDLy brings new features like 3D convolutions and custom hexagonal kernels but also has a slightly changed API to be more PyTorch-like.**
Changes
Dependencies
This version of HexagDLy requires Python >= 3.5 and was tested to work with PyTorch v0.4.1.
Syntax
The syntax has changed to follow torch.nn with capitals in the class names. A convolution e.g. is now defined with `hexconv = hexagdly.Conv2d()`.
3D Convolutions
3 dimensional convolutions can be performed via `hexagdly.Conv3d()` where the hexagonal plane is indexed with the last two indices of the input tensor.
Custom Kernels
With the implemented `hexagdly.Convd_CustomKernel()` hexagonal kernels of 2 and 3 dimensions can be filled with arbitrary values via numpy arrays for each sub-kernel.