v1.0.0

Additions

• Added support for 3D inputs in convolutional and pooling layers.
• Added shape, scale, and mode parameters to Upsample.
• Added channels parameter to AdaptivePool to specific if channels dimension exists.
• Added support for multiple strides in convolutional and pooling layers.
• Added dropout parameter for all types of convolutional layers.
• Added scale parameter to all scaling layers.
• Added PixelShuffle layer.
• Added support for multiple kernel and padding sizes in pooling layers.
• Added mode parameter for all pooling layers.
• Added checkpoint parameter to Composite layer.
• Added more warnings and errors to catch problems earlier.

Changes

• Changed all layers from functions to classes.
• Changed PixelShuffle1d to PixelShuffle for N-dimensional pixel shuffling.
• Changed config_path parameter to config_dir and name in Composite layer.
• Changed Composite layer to be more independent of Network.

Fixes

• Fixed kl_loss Network attribute not being sent to device.

Removals

• Removed defaults.py, defaults are now built into the classes.
• Removed latent_mse_weight attribute as it was unneeded.
• Removed 2d parameter from network and layers, it is now implicitly worked out.