NeuraLisp is a machine learning framework for Common Lisp, designed for ease of use and extensibility. It provides a modular implementation for building, training, and evaluating neural networks for various machine learning tasks.
- Core components for tensor operations, autograd, GPU support, and more
- A variety of layers, including linear, convolutional, recurrent, and attention-based layers
- Common activation functions like ReLU, Sigmoid, and Tanh
- Loss functions like mean squared error and cross-entropy
- Optimizers like Stochastic Gradient Descent (SGD) and Adam
To install NeuraLisp, clone the repository and load the system definition using Quicklisp or ASDF:
git clone https://github.com/yourusername/NeuraLisp.git
Then, in your Common Lisp REPL:
(ql:quickload :NeuraLisp)Here's an example of how to create a simple neural network using NeuraLisp:
(use-package :NeuraLisp.core.tensor)
(use-package :NeuraLisp.layers.linear)
(use-package :NeuraLisp.activations.relu)
;; Create a linear layer
(defvar *layer* (make-instance 'linear :input-dim 3 :output-dim 2))
;; Create input tensor
(defvar *input* (make-tensor #(1.0 2.0 3.0) #(1 3)))
;; Apply the linear layer
(defvar *output* (forward *layer* *input*))
;; Apply ReLU activation
(defvar *relu* (make-instance 'relu))
(defvar *activated-output* (activate *relu* *output*))For more examples, please refer to the examples/ folder.
Detailed documentation for each component can be found in the docs/ folder.
We welcome contributions to NeuraLisp! If you find a bug, want to improve the code quality, or have ideas for new features, please feel free to create an issue or submit a pull request on GitHub.
NeuraLisp is licensed under the MIT License. Please see the LICENSE file for more information.