Python/NumPy is great for numerical computing! For larger neural networks, however, GPU accelerated array operations are needed for decent run times. CUDArray tries to combine the easy of development from the NumPy with the computational power of Nvidia GPUs in a lightweight and extensible framework. While CUDArray only implements a small subset of NumPy it is mature enough to support a neural network pipeline as demonstrated in the deeppy project.
CUDArray is an open source project under the MIT license. Source and installation instructions are available from the Github repository.