javascript - Brain.js 倒数/反转训练(根据输出预测输入)

Question

我刚刚学会了训练 brain.js network 并且只是在玩它。然后我很好奇是否可以采取相反的方式 - 从输出预测输入？

这是我的代码

const brain = require('brain.js');
const network = new brain.NeuralNetwork();    
/* 
network.train([
    { input: { doseA: 0 }, output: { indicatorA: 0 } },
    { input: { doseA: 0.1 }, output: { indicatorA: 0.02 } },
    { input: { doseA: 0.2 }, output: { indicatorA: 0.04 } },
    { input: { doseA: 0.3 }, output: { indicatorA: 0.06 } },
    { input: { doseA: 0.4 }, output: { indicatorA: 0.08 } },
    { input: { doseA: 0.5 }, output: { indicatorA: 0.10 } },
    { input: { doseA: 0.6 }, output: { indicatorA: 0.12 } },
    { input: { doseA: 0.7 }, output: { indicatorA: 0.14 } },
], {
  iterations: 1e6,
  errorThresh: 0.00001
});
*/    
network.fromJSON({"sizes":[1,3,1],"layers":[{"doseA":{}},{"0":{"bias":-0.7720749378204346,"weights":{"doseA":-6.819720268249512}},"1":{"bias":0.2317514568567276,"weights":{"doseA":-1.4340121746063232}},"2":{"bias":-0.34450986981391907,"weights":{"doseA":-2.9449453353881836}}},{"indicatorA":{"bias":-1.0124520063400269,"weights":{"0":-5.02399206161499,"1":-1.69333016872406,"2":-3.1710503101348877}}}],"outputLookup":true,"inputLookup":true,"activation":"sigmoid","trainOpts":{"iterations":1000000,"errorThresh":0.00001,"log":false,"logPeriod":10,"learningRate":0.3,"momentum":0.1,"callbackPeriod":10}})

const result = network.run({ doseA: 0.35 });   
console.log(result);
// { indicatorA: 0.06978786736726761 }

但是如果我愿意怎么办network.run({ IndicatorA: 0.07 }) 并得到 {doseA: 0.35 }?

我是否被迫再次训练网络，但切换了输入和输出？或者有什么办法可以逆转吗？

Answer 1

绝对有一个实用程序可以完成这件事:https://github.com/BrainJS/brain.retro.js

如果有更多兴趣，我还可以获得查询实用程序，它允许您执行一些非常有趣的 mongo 风格查询:https://github.com/BrainJS/brain.retro.js/blob/master/query.js