nn.el Building and Training Neural Networks in Emacs Lisp

1 Overview

nn.el is a zero dependancy emacs lisp library to build and train neural networks.

2 Installation

Firstly clone the directory, and then run make to build the emacs modules matrix.c and ops.c. Then add the directory to your load path with something like

(add-to-list 'load-path "~/where/you/cloned")

3 Usage

Below is an example on how to train a (relatively large) neural network on some randomly generated data

(require 'nn)
(require 'matrix)
(require 'ops)

(setq model `(,(nn-layer 100 80 'ops-relu)
              ,(nn-layer 80 10 'ops-relu)
              ,(nn-layer 10 8 'ops-relu)
              ,(nn-layer 8 3)))


(setq x (matrix-random 100 2))
(setq y (matrix-transpose (vconcat [[1 0 0]] [[0 0 1]])))

(let* ((s (ops-softmax (nn-forward-layers x model)))
      (loss (nn-crossentropy y s)))
  (message "Initial loss: %s" loss))
         


(defun nn--apply-gradient-sgd-layer (grad layer)
  "Apply GRAD to LAyER using gradient descent"
  (let ((w (nth 0 layer))
        (b (nth 1 layer))
        (wg (nth 0 grad))
        (bg (nth 1 grad)))
    (setf (nth 0 layer) (matrix-subtract w (matrix-scalar-mul 0.01 wg)))
    (setf (nth 1 layer) (matrix-subtract b (matrix-scalar-mul 0.01 bg)))))

(dotimes (counter  100)
  ;; this is the train step; essentially just do this as many times
  (let ((grads (nn-gradient x y model)))
    (seq-mapn #'nn--apply-gradient-sgd-layer grads model)))

(let* ((s (ops-softmax (nn-forward-layers x model)))
      (loss (nn-crossentropy y s)))
  (message "after training loss: %s" loss))