PyTorch implementation of Geoffrey Hinton's Forward Forward Algorithm (https://www.cs.toronto.edu/~hinton/FFA13.pdf)

• Save model
• Save checkpoints
• Add pytorch decorators where needed
• Restart training from checkpoints if exist
• Check GPU usage on device move
• Add RNG environment class
• Add Non-Greedy Trainer
• Change train and "validation" logging for NonGreedy to measure at the end of each pass.
• Current embed labels has an issue where by chance it might assign a "negative" sample with the "correct" label.
• Add classes for Goodness
• Add classes for Metrics
• Comment code
• Add preamble comment section
• Restart step metric with every layer. Currently continues to increment layer by layer.
• Early stopping at layer epoch level
• Add reference to PyTorch and Mohammed's implementation
• Add softmax classifier as easy predict
• Move current classifier predict to hard predict
• Add test config
• Load model from location
• Add test script
• Add W&B logging
• Add GLOM model
• Unsupervised case
• Experimentation (threshold)
• Experimentation (goodness function)
• Experimentation (AE GLOM)
• Improve W&B Logging (Gradients)
• Improve W&B Logging (Example of activations)
• Improve W&B Logging (Gradients)
• Improve W&B Logging (Config)
• Improve W&B Logging (Tags)
• Improve W&B Logging (Groups)
• Improve W&B Logging (Charts config - Log Scale, axis labels etc)