forward method

ValueVector forward(ValueVector input)

Implementation

ValueVector forward(ValueVector input) {
  if (input.values.length != features) {
    throw ArgumentError(
        "Input vector size (${input.values.length}) must match features ($features)");
  }

  // Calculate mean
  Value sum = input.values.reduce((a, b) => a + b);
  Value mean = sum / Value(features.toDouble());

  // Calculate variance (using data, not Value objects for intermediate step to avoid graph explosion)
  // For proper backprop, these intermediate operations should also create Value objects
  // This is a simplification; a full implementation would build these into the computational graph.
  Value varianceSum = Value(0.0);
  for (Value val in input.values) {
    varianceSum += (val - mean).pow(2);
  }
  Value variance = varianceSum / Value(features.toDouble());

  // Normalized input
  List<Value> normalizedValues = [];
  for (Value val in input.values) {
    normalizedValues.add((val - mean) / (variance + Value(epsilon)).sqrt());
  }
  ValueVector normalizedInput = ValueVector(normalizedValues);

  // Apply learned scale and shift
  List<Value> outputValues = [];
  for (Value val in normalizedInput.values) {
    outputValues.add(gamma * val + beta);
  }
  return ValueVector(outputValues);
}