main function

void main()

Implementation

void main() {
  const lr = 0.01;
  final model = MultiLayerPerceptron(lr); // 4 inputs → 2 outputs
  print("Image bytes legnt: ${imgBytes.length}"); // 784 bytes
  final inputs = [
    ValueVector.fromDoubleList(imgBytes
        .map((e) => e / 255.0)
        .toList()), // Correct casting might be needed
    ValueVector.fromDoubleList(imgBytes.map((e) => e / 255.0).toList()),
    ValueVector.fromDoubleList(imgBytes.map((e) => e / 255.0).toList())
  ];

  final targets = [
    ValueVector([
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(1.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0)
    ]),
    ValueVector([
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(1.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0)
    ]),
    ValueVector([
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(1.0),
      Value(0.0),
      Value(0.0),
      Value(0.0),
      Value(0.0)
    ])
  ];

  const epochs = 400;

  for (int epoch = 0; epoch < epochs; epoch++) {
    final losses = <Value>[];

    // Reset gradients

    // Compute loss for all samples
    for (int i = 0; i < inputs.length; i++) {
      final yPred = model.forward(inputs[i]);
      final yTrue = targets[i];
      final diff = yPred - yTrue;
      final squared = diff.squared();
      final sampleLoss = squared.mean();
      losses.add(sampleLoss);
    }

    final totalLoss = losses.reduce((a, b) => a + b);
    // final avgLoss = totalLoss * (1.0 / inputs.length);
    // avgLoss.backward();
    totalLoss.backward();

    // Gradient descent
    model.updateWeights();

    // if (epoch % 4 == 0) {
    //   print("Epoch $epoch | Loss = ${totalLoss.data.toStringAsFixed(4)}");
    // }
    if (epoch % 4 == 0) {
      print("Epoch $epoch | Loss = ${totalLoss.data.toStringAsFixed(10)}");
    }
    model.zeroGrad();
  }

  for (var input in inputs) {
    // Reset gradients
    for (var p in model.parameters()) {
      p.grad = 0;
    }
    // print("Input: ${input}");
    print("Output: ${model.forward(input)}");
    print("");
  }
}