Library to build, differentiate, and optimize computational graphs. Supports differentially private optimization via mixins.
Install dependencies with:
pip install -r requirements.txt
| Folder | Name |
|---|---|
| Environments | Classes for wrapping data |
| Neural_Network | Node implementations for building a computational graph |
| Tests | Example networks, compatible with pytest |
- Optimizers
- Gradient Descent
- Momentum
- Nesterov Accelerated Gradient
- Adagrad
- RMSProp
- Adam
- Adamax
- Nadam
- Quickprop
- Selectable functions
- Basis / Activation
- Regression (softplus, tanh, etc.)
- Classification (use softmax on final layer)
- Cost (sum squared and cross entropy)
- Basis / Activation
- Network Configuration
- Construct flexible graphs of vector functions
- Multiple layers
- Set number of nodes for each layer
- Optimizer Configuration
- Batch size (set to one for stochastic descent)
- Set hyperparameters for each layer, or broadcast one
- Learning rate
- Parameters specific to convergence algorithm
- Multiprocessed graphing
This is a rewrite of a more rigid 'caterpillar' network: https://github.com/Shoeboxam/Neural_Network