我试图用brain.js来预测未来的销售额,但我不知道我做错了什么。
我正在跟踪this tutorial,它在那里工作。
我有最新版本的brain.js,我找不到其他问题的这种性质。
另外,你也可以提出一个更好的预测方法。
- 错误:**
/node_modules/brain.js/src/recurrent/rnn-time-step.js:279
const result = [lastOutput.weights[0]];
^
TypeError: Cannot read property 'weights' of undefined
at LSTMTimeStep.forecastNumbers (/node_modules/brain.js/src/recurrent/rnn-time-step.js:279:32)
at LSTMTimeStep.runObject (/node_modules/brain.js/src/recurrent/rnn-time-step.js:293:14)
at LSTMTimeStep.run (/node_modules/brain.js/src/recurrent/rnn-time-step.js:104:21)
at Object.<anonymous> (/forecast.js:123:17)
- 代码:**
const brain = require("brain.js");
let data = [
{ sales: 0 },
{ sales: 0 },
{ sales: 0 },
{ sales: 0 },
{ sales: 0 },
{ sales: 0 },
{ sales: 92 },
{ sales: 759 },
{ sales: 3691 },
{ sales: 4039 },
{ sales: 2257 },
{ sales: 1736 },
{ sales: 3979 },
{ sales: 3170 },
{ sales: 6092 },
{ sales: 7839 },
{ sales: 5764 },
{ sales: 5512 },
{ sales: 5494 },
{ sales: 7458 },
{ sales: 3721 },
{ sales: 8512 },
{ sales: 1089 },
{ sales: 7462 },
{ sales: 710 },
{ sales: 4534 },
{ sales: 6224 },
{ sales: 7610 },
{ sales: 3976 },
{ sales: 6243 },
{ sales: 1532 },
{ sales: 2204 },
{ sales: 801 },
{ sales: 1575 },
{ sales: 2144 },
{ sales: 3679 },
];
let max = Math.max(...data.map(o => o.sales));
let min = Math.min(...data.map(o => o.sales));
function normalize(step) {
return { sales: (step.sales - min) / (max - min) };
}
let scaledData = data.map(normalize)
let trainingData = [
scaledData.slice(0, 12),
scaledData.slice(12, 24),
scaledData.slice(24, 36),
]
const net = new brain.recurrent.LSTMTimeStep({
inputSize: 1,
hiddenLayers: [1],
outputSize: 1
});
net.train(trainingData, {
iterations: 200,
learningRate: 0.005,
errorTresh: 0.02
})
console.log(net.run(trainingData[0]));
2条答案
按热度按时间gz5pxeao1#
嘿,这可能是brainiderjs的问题,我也有类似的问题
TypeError: Cannot read properties of undefined (reading 'rows')
ubof19bj2#
在brain.js github文档中,有一节“使用RNNTimeStep、LSTMTimeStep和GRUTimeStep进行培训”:
对于**“LSTMTimeStep”**,每个训练模式可以:
所以你可以试试
normalize
函数将只返回一个数字,scaledData
将是一个数字数组。希望这能有所帮助。