Micrograd-Go

Micrograd-go is an simple implementation of the Micrograd library in Go. It contains the same primitives as the original library along with a small neural network library on top. In theory this is enough to train up deep neural networks for regression and classification based tasks (as shown in the train.go).

Usage

Running the Training Loop

The example in train.go shows a full e2e example of a classification task using a multi-layer perceptron (MLP). To run the training loop, use the following command:

go run train.go

You can enable verbose output with the -verbose flag:

go run train.go -verbose

Tests

Both the engine and nn packages have unit tests. To run the tests for the engine and neural network packages, use:

go test ./...

To visualize the DAG of a neural network, you can use the Test_Diagram function in engine/engine_test.go.

Structure

.
├── engine/
│   ├── engine.go
│   ├── engine_test.go
    └── graph.go
├── nn/
│   ├── nn.go
│   └── nn_test.go
├── README.md
├── train.go
├── go.mod
└── go.sum

The engine package contains the core primative of Micrograd-Go: a Value which is a scalar value with a gradient. We implement mathematical operations on Value object (which gets a bit messy due to Go's lack of operator overloading). The nn package provides a small neural network libary with support for the Neuron, Layer, and MLP types. The graph.go file contains a helper function to draw the DAG (built using @tmc's dot library).

Using the Library in Your Project

To use the library in your own Go project, you can import it as follows:

import (
    "github.com/ishandhanani/micrograd-go/engine"
    "github.com/ishandhanani/micrograd-go/nn"
)

You can create a simple MLP as follows:

mlp := nn.NewMLP(3, []int{4, 4, 1})

This will create an MLP with 3 input neurons, 2 hidden layers of 4 neurons each, and 1 output neuron.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License

Acknowledgements

This project is based on Andrej Karpathy's micrograd. Check out his YouTube video for an in-depth explanation of the concepts.