Scala.js Facade Strategies Example

This workspace showcases two distinct methodologies for interacting with JavaScript libraries from Scala.js 3, using TensorFlow.js (@tensorflow/tfjs v4.22.0) as the underlying engine.

It has been structured as a multi-project aggregate SBT build, splitting the strategies into two modules:

1. handcrafted: Manually designed, type-safe facade encapsulated inside a package-private boundary, exposing a 100% pure-Scala API. Trains a Feedforward Neural Network to solve the classic non-linear XOR gate problem.
2. scalablytyped: Automatically generates extensive, ultra-precise Scala.js facades from TypeScript definition files (.d.ts) using the ScalablyTyped toolchain.

---

Prerequisites

To compile and run both subprojects, you need:

• Java JDK (recommended: JDK 11, 17, or 21)
• SBT (Scala Build Tool)
• Node.js (v18+) and npm

---

1. Handcrafted Module (Neural Network XOR)

This submodule demonstrates the Private Raw Facade + Pure Scala Wrapper architectural pattern. Inside Network.scala, we define package-private (private[network]) raw JS facades mapping directly to @tensorflow/tfjs imports and classes. The public Network class acts as a clean barrier, wrapping all JS promises into Scala Futures, translating nested standard Scala sequences into multi-dimensional typed JS arrays, and automatically managing C++ tensor allocations via .dispose() to ensure zero memory leaks.

How to Run:

Launch the handcrafted XOR neural network directly via SBT:

sbt "handcrafted/run"

What it does:

1. Builds a Feedforward Neural Network using sequential layers (ReLU hidden layer, Sigmoid output layer).
2. Translates pure Scala datasets (inputs and outputs lists) into 2D tensors.
3. Trains the model over 1000 epochs using the adam optimizer and binaryCrossentropy loss.
4. Performs predictions on the standard inputs, showing the trained XOR output (approaching 0 and 1 correctly).
5. Disposes of all input, output, and prediction tensors.

---

2. ScalablyTyped Module (Tensor Squaring)

This submodule leverages the ScalablyTyped sbt plugin to parse @tensorflow/tfjs type definitions and automatically output Scala.js types under the custom package customtypings. It uses a custom webpack.config.js specifying target node.

How to Run:

You can run the automatic facade generation, Webpack bundling, and execution in one single command without setting any shell environment options (like NODE_OPTIONS):

sbt -J-Xmx4G "scalablytyped/run"

What it does:

1. Imports and utilizes auto-generated facades from customtypings.tensorflowTfjs.mod and associated core math libraries.
2. Creates a 1D tensor representing elements [1.0, 2.0, 3.0, 4.0].
3. Performs a typed squaring operation calling the standalone square function from customtypings.tensorflowTfjsCore.distOpsSquareMod.
4. Outputs the original and squared tensors via native .print() calls inside the Node console.
5. Safely deallocates tensor memory via .dispose().