A ML library in C++.

Why FJML?

• Written completely from scratch
  • Implemented in an easy-to-understand way
  • Completely self-contained, does not rely on libraries

Both low level and high level

• Written in C++
• Includes low level components
• Includes high level abstractions

Flexible and easily extensible

Installation

Run make init, then make, and then sudo make install.

To compile programs with FJML, just use g++ myfile.cpp -lFJML

Why Farmer John

Elements of this library were inspired by USACO. The library is written with simple code, in the style that most programmers with little experience outside of USACO can understand.

Features

Currently has support for:

• Activations:
  • Sigmoid
  • Linear
  • ReLu
  • Swish
  • Leaky ReLu
  • Tanh
• Layers:
  • Dense layers
  • Softmax layers
• Loss Functions:
  • Mean Squared Error
  • Huber
  • Log Loss (Cross Entropy)
• Optimizers:
  • SGD
  • Adam