Make sure conditioning stack is properly trained

Changes for paper revision

Update README.md

Fixed passing sample_shape to setup_data - Issue #9

Add option to continue autoencoder training from checkpoint - Issue #7

Fixed and verified evaluation scripts - Issue #5

Fix import in train_nowcaster.py - Issue #4

Fix to enable demo to run on CPU only

Fixed bug in forecast demo with one ensemble member

Added missing .gitignore to repository

Update README.md

Added multi-GPU ForecastDistributed class

Add preprint info to README.md

Added split data file - Issue #1

Update README.md