Chapel implementation of GCP tensor decomposition. The initial implementation was by Richard Barrett (Sandia National Labs).
Intial implementation supports a 3-mode tensor composed of Poisson distributed count data, using Stochastic Gradient Descent. The model tensor is constructed using all tensor elements, i.e. is not sampled.
The modular design should enable extension to other distributions, optimization methods, and loss functions.
'count' data 'sgd' optimization.
Sampling not yet implemented.