NNKit: A dynamic neural network framework.
Project description
# NNKit: A dynamic neural network framework.
NNKit is a framework for training neural network models.
It is based on the concept of dynamic computation graphs, where graphs are implicitly generated on each forward pass as nodes are passed into other nodes as inputs
This dynamic approach makes it practical to implement recurrent topologies for varying length inputs and allows generating graphs with a minimal amount of code.
#### Modules The following is a list of net nodes and optimizers in the framework, along with the framework version in which they were added.
## activation: * ReLU (1.0) * LReLU (1.0) * Sigmoid (1.0) * Tanh (1.0) * Softmax (1.0)
## arithmetic: * Multiply (1.0) * Add (1.0)
## loss: * L1 (1.0) * L2 (1.0) * Cross Entropy (1.0)
## normalization: * Batch Normalization (1.0)
## regularization: * L2 (1.0) * Dropout (1.0)
## training: * Gradient descent w/ momentum (1.0) * Adam / RMSProp (1.0)
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.