Getting Started

Welcome to Gradien. This guide will help you set up the library and train your first neural network.

Installation

Method 1: Wally Recommended

If you use Wally for dependency management, add Gradien to your wally.toml.

wally.toml

[dependencies]
Gradien = "eternitydevs/gradien@1.4.0-rc5"

Terminal

wally install

Method 2: RBXM

1. Download the latest release from the GitHub Releases page.
2. Drag and drop the .rbxm file into ReplicatedStorage in Roblox Studio.

---

Your First Training Loop

Let's create a simple model that learns to approximate a linear function.

Step 1: Setup

Require the library and define a basic dataset.

local Gradien = require(game.ReplicatedStorage.Gradien)

-- Create dummy data: Inputs (X) and Targets (Y)
-- Shape: {Features=2, Batch=5}
local X = Gradien.Tensor.fromArray({
    1, 2, 3, 4, 5, 
    1, 2, 3, 4, 5
}, {2, 5}) 

local Y = Gradien.Tensor.fromArray({
    2, 4, 6, 8, 10
}, {1, 5})

Step 2: Define Model

We will use a Sequential container with Linear layers and ReLU activations.

local model = Gradien.NN.Sequential({
    Gradien.NN.Linear(2, 8),             -- Input -> Hidden
    function(_, x) return Gradien.NN.Activations.ReLU(x) end,
    Gradien.NN.Linear(8, 1)              -- Hidden -> Output
})

Step 3: Trainer

Use the Trainer class to handle the training loop, backpropagation, and optimization automatically.

local trainer = Gradien.Trainer.new({
    model = model,
    optimizerFactory = function(params) 
        return Gradien.Optim.Adam(params, 1e-2) 
    end,
    loss = Gradien.NN.Losses.mse_backward,
    metric = function(pred, target)
        -- Custom metric logic here
        return 0
    end
})

-- Start training for 100 epochs
trainer:fit(function()
    -- Dataloader generator
    return function() return X, Y end
end, { epochs = 100 })