gograd is a lightweight and simplified implementation of a deep learning library in Go. It is inspired by the original TinyGrad library and provides a minimalistic yet functional framework for building and training neural networks.
- TODO: Backpropagation: Automatic differentiation for computing gradients of the loss function with respect to the model parameters.
- TODO: Optimization: Stochastic Gradient Descent (SGD) optimization for updating the model parameters during training.
- TODO: Dense Layer: Implementation of a fully connected layer that can be used to build feedforward neural networks.
- TODO: Activation Functions: Support for common activation functions like ReLU, Sigmoid, and Tanh.
- TODO: Loss Functions: Built-in loss functions such as Mean Squared Error (MSE) and Cross-Entropy.
- TODO: Add Batch Normalization for improved training performance and stability.
To use gograd, you need to have Go installed on your system. If you don't have it, you can download it from the official website: https://golang.org/
Once you have Go installed, you can install gograd using go get:
go get github.com/campv/gogradContributions to gograd are welcome! If you have suggestions, bug reports, or improvements, please open an issue or submit a pull request.
gograd is licensed under the MIT License.
This project is inspired by the original micrograd library micrograd library and tinygrad library