can now condition on psd and load random psds
Series of changes:
Model
- Can now be conditioned on a psd
- Can now separate channels so that each are passed through a separate network
- Added some extra options for layers
- Transformer
- Seperable Conv
Config
- Updated config defaults with extra parameters and model specification
Loading data
- Can now randomise the PSD for each piece of data when loaded #6 (closed)
- Can also input the PSD to condition network on PSD
- MAJOR bug fix in data augmentation, issue #5 (closed)
- when augmenting distance it was using uniform distributions so now the distance prior has changes this modifies the prior distribution, fixed to resample from the prior
Training
- Fixed the ramp callback was not working before, issue #7 (closed)