minigrad

A minimal implementation of autograd and neural networks. Inspired by micrograd and pytorch. This library is meant to be educational and not meant for production code.

Install and dev work:

Install: go get github.com/NoahSchiro/minigrad@latest

Import in your code:

// You may only need one of these depending on your work
import "github.com/NoahSchiro/minigrad/tensor"
import "github.com/NoahSchiro/minigrad/ndarray"
import "github.com/NoahSchiro/minigrad/nn"

Dev work:

Install: git clone [email protected]:NoahSchiro/minigrad.git

Run tests: go test ./...

Example:

Currently, there is one example for the xor problem which can be found in cmd/xor.go

Features and roadmap:

• Autograd engine
• Linear layers
• RNN, LSTM (targeting v2.1)
• CNN (targeting v2.2)
• Sigmoid
• ReLU
• SoftMax (targeting v2.1)
• SGD optimizer
• Adam optimizer (targeting v2.1)

SCC Report

The goal is to keep the library under 1k lines of code. This excludes test files.

Language	Files	Lines	Blanks	Comments	Code
Go	12	1016	163	74	779