SOURCE

// 核心输入数据
const datas = [
  [[0,0], 0],
  [[0,1], 0],
  [[1,0], 0],
  [[1,1], 0],
];

// 随机权重数组
const weights = [];
for (let i=0; i<3; i++){
  weights[i]=Math.random()-0.5;
}

// 激活
function calcOutput(inputs){
  let output = inputs[0]*weights[0] + inputs[1]*weights[1] + 1*weights[2];
  return sigmoid(output);
}

// 计算误差大小
function errRate(output, excepted){
  return Math.abs(output - excepted);
}

// 计算最近n次的错误率的平均值
const errors = [];
const maxError = 20;
function calcError(err){
    errors.push(err);

    if(errors.length>maxError){
        errors.shift();
    }

    return errors.reduce((tmp,item)=>tmp+item)/errors.length;
}

// 大数映射成0~1的数
function sigmoid(x){
  return 1/(1 + Math.pow(Math.E, -x));
}

// 试着调整每个权重
const d = 0.000001;
const trainRate = 0.1;
let times = 0;
const threshold = 0.001;
function train(inputs, excepted){
  let err = errRate(calcOutput(inputs), excepted);
  
  const dw = [];
  weights.forEach((w, i) =>{
    weights[i] += d;
    
    let err2 = errRate(calcOutput(inputs), excepted);
    
    dw[i] = (err2 - err)/d;
    
    weights[i] = w;
  });
  
  weights.forEach((w, i)=>{
    weights[i] -= dw[i]*trainRate;
  });
  
  let e=calcError(err);
  times++;

  if(times%500==0){
    console.log(`#${times} ${e}`);
  }

  return e<=threshold;
}

for(let i=0;;i++){
    let data = datas[i%datas.length];

    if(train(data[0], data[1])==true){
     break;
    };
}

console.log(weights);