Redesigned for more powerful networks (now supports autoencoders as well). Very easy to write small models (inspired by keras).
Features:
- Layers: Dense, Dropout, Flatten
- Activators: Sigmoid, Tanh, Relu
- Loss Functions: Quadratic, Hillinger, Cross-entropy
- Variable Initializers: Zeros, Normal, Xavier
- Optimizers: SGD, Momentum, RMSProp, AdaGrad, Adam