primitiv Documentation

primitiv is a neural network library developed by National Institute of Information and Communications Technology (NICT) and Nara Institute of Science and Technology (NAIST). primitiv is written in C++11 and supports some other languages such as Python and Rust through bindings. primitiv allows users to write their own networks using a define-by-run style construction methods, and most features in the library is designed device-independent. Users can perform the own network using various computing backends such as Eigen, CUDA and OpenCL with no (or few) modifications of the original code.

Contents:

• 1. Installing primitiv
• 2. primitiv C++ Tutorials
  • 2.1. Step-by-step Example: Solving the XOR Problem
• 3. Library Designs
  • 3.1. Headers, Library Files and Compilation
  • 3.2. Shapes and Operation Rules
  • 3.3. Nodes and Tensors
  • 3.4. Devices
  • 3.5. Default Device and Graph
  • 3.6. Training
  • 3.7. Models
  • 3.8. Saving and Loading
• 4. primitiv Reference
  • 4.1. primitiv API Reference
    • 4.1.1. Devices
    • 4.1.2. Functions
    • 4.1.3. Graph
    • 4.1.4. Initializers
    • 4.1.5. Model
    • 4.1.6. Node
    • 4.1.7. Optimizers
    • 4.1.8. Parameter
    • 4.1.9. Shape
    • 4.1.10. Tensor
  • 4.2. Build Options
  • 4.3. primitiv File Format v0.1