MLGodotKit

Empower your Godot projects with the power of machine learning!
MLGodotKit is a C++ GDExtension for Godot, enabling seamless integration of AI-driven features into your games and applications. With support for adaptive behaviors and real-time decision-making, it’s designed to inspire innovation and enhance gameplay.

🎉 New Feature: WebAssembly Support

WebAssembly (WASM) binaries have been added to the project to enable web-based game development.
You can now build and run your game directly in the browser using Godot's Web export.

Play the Web Demo:
Launch Game in Browser

📁 The exported web build is located at:
examples/NNPong/web/

📘 Learn how to compile GDExtensions for the Web:
Godot Docs – Compiling for Web

Node Extensions in Progress

• Data Science Tooling: Tools for preprocessing, analyzing, and visualizing data directly in Godot.
• Deep Learning Integration: Simplified nodes for building and deploying neural networks.
• Cross-Platform Compatibility: Fully supported on Windows, Linux, and macOS.

---

LIMITATIONS

Please note that this is a work in process pet project of mine and as such will come with a few bugs. If you enjoy the project and want to support please open an issue for desired features or bug fixes!

Next Features to be added:

• A linear layer for the Neural Network Model for regression based predictions
• A robust collection of popular loss functions for out of the box use cases
• More supervised and unsupervised model implementations
• If you'd like to see something else add an issue!

Getting Started

Here are quick examples of how to use the three core models in MLGodotKit:

All models are GDExtension nodes and can be used directly in any Godot scene or script.

---

Linear Regression (LRNode)

@onready var lr_model = LRNode.new()

func _ready():
	lr_model.set_learning_rate(0.01)
	lr_model.initialize(1)

	# Simple linear dataset: y = 3x + 5
	var inputs = [[1], [2], [3]]
	var targets = [[8], [11], [14]]

	lr_model.train(inputs, targets, 1000)

	var prediction = lr_model.predict([4])
	print("Predicted y for x=4:", prediction)

---

Descision Tree Classifier (DTreeNode)

@onready var tree = DTreeNode.new()

func _ready():
	tree.set_max_depth(5)
	tree.fit([[1], [2], [3], [10], [11], [12]], [0, 0, 0, 1, 1, 1])

	var result = tree.predict([[2], [11]])
	print("Predictions:", result)  # [0, 1]

---

Neural Network (NNNode)

Note: All loss functions are left to the user to implement please see examples for suggested implementation will be implemented natively at a later date.

@onready var nn = NNNode.new()

func _ready():
	nn.set_learning_rate(0.1)
	nn.add_layer(2, 4, "relu")
	nn.add_layer(4, 1, "sigmoid")

	var inputs = [[0,0], [0,1], [1,0], [1,1]]
	var targets = [[0], [1], [1], [0]]  # XOR logic

	for i in range(5000):
		for j in range(inputs.size()):
			var y_pred = nn.forward([inputs[j]])[0]
			var error = y_pred - targets[j][0]
			nn.backward([[2.0 * error]])

	print("Test XOR:")
	for i in range(inputs.size()):
		var output = nn.forward([inputs[i]])
		print("Input:", inputs[i], " Predicted:", output, "Expected:", targets[i])

---

Credits

Built on the powerful Eigen C++ library.