PyTorch implementation of DoppelGANger. This library has been moved to the gretel-synthetics package. Find the latest version here: https://github.com/gretelai/gretel-synthetics/tree/master/src/gretel_synthetics/timeseries_dgan
Using the wiki data from the original paper, we expect to see a 7-day pattern in autocovariance of generated data. Instead, when training for 400 epochs there's almost no correlation.
Attribute distributions match pretty well between real and generated:
So seems like something specifically wrong in the generation of features.