This extends the current implementation of QRNN to support both Pytorch and Keras
as backends. The QRNN class has been refactored to separate general QRNN functionality
from the underlying neural network model. Some functionality has been added e.g. computing
the posterior mean and sampling from a fitted Gaussian.
This work is part of an upcoming study on cloud correction at 183 GHz.
This extends the current implementation of QRNN to support both Pytorch and Keras as backends. The QRNN class has been refactored to separate general QRNN functionality from the underlying neural network model. Some functionality has been added e.g. computing the posterior mean and sampling from a fitted Gaussian.
This work is part of an upcoming study on cloud correction at 183 GHz.