class NeuralNetwork {

  constructor(inputNodes, hiddenNodes, outputNodes) {
    this.inputNodes = inputNodes;
    this.hiddenNodes = hiddenNodes;
    this.outputNodes = outputNodes;

    this.weights = new Array(inputNodes * hiddenNodes);
    this.biases = new Array(hiddenNodes);

    for (var i = 0; i < inputNodes * hiddenNodes; i++) {
      this.weights[i] = Math.random();
    }

    for (var i = 0; i < hiddenNodes; i++) {
      this.biases[i] = Math.random();
    }
  }

  train(inputs, outputs) {
    // Forward pass
    var hiddenOutputs = new Array(hiddenNodes);
    for (var i = 0; i < inputNodes; i++) {
      for (var j = 0; j < hiddenNodes; j++) {
        hiddenOutputs[j] += inputs[i] * this.weights[i * hiddenNodes + j];
      }
    }

    for (var j = 0; j < hiddenNodes; j++) {
      hiddenOutputs[j] += this.biases[j];
    }

    // Backpropagation
    var error = new Array(outputNodes);
    for (var i = 0; i < outputNodes; i++) {
      error[i] = outputs[i] - hiddenOutputs[i];
    }

    for (var j = 0; j < hiddenNodes; j++) {
      for (var i = 0; i < inputNodes; i++) {
        this.weights[i * hiddenNodes + j] += error[j] * hiddenOutputs[j] * inputs[i];
      }
    }

    for (var j = 0; j < hiddenNodes; j++) {
      this.biases[j] += error[j] * hiddenOutputs[j];
    }
  }

  predict(inputs) {
    // Forward pass
    var hiddenOutputs = new Array(hiddenNodes);
    for (var i = 0; i < inputNodes; i++) {
      for (var j = 0; j < hiddenNodes; j++) {
        hiddenOutputs[j] += inputs[i] * this.weights[i * hiddenNodes + j];
      }
    }

    for (var j = 0; j < hiddenNodes; j++) {
      hiddenOutputs[j] += this.biases[j];
    }

    // Output
    var outputs = new Array(outputNodes);
    for (var j = 0; j < outputNodes; j++) {
      outputs[j] = hiddenOutputs[j];
    }

    return outputs;
  }
}