- Significantly improve Animator API
- Fix temporal encoding being unused
- Parameterize average pooling size in classification with MLP head
- Let user supply class label names to classifier
- Classification visual in tensorboard: Show bar plot of softmax
- Separation of perception, rule and head model in BasicNCAModel for improved modularity
- Classification: Average pooling between NCA backbone and classification head
- Dependency updates
- Moved training/inference step parameter from Trainer to BasicNCAModel
- Fix use of max() reduction in uncertainty estimator
- Add uncertainty estimation feature
- Rename VisualMultiImageClassification --> VisualRGBImageClassification
- Deprecate VisualMultiImageClassification
- Add record() member to BasicNCAModel to record sequences of predictions
- Adapt Animator class to new prediction API
- Use Animator class in growing emoji example task
- Fix evaluation script for MNIST example task
- Fix dimensionality error in "classify" method
- Add evaluation scripts for MedMNIST example tasks
- Fix visualization of normalized data
- Add (optional) MLP classification head in classification model
- Slightly adjust training parameters (learning rate and adam betas)