tensorflow.js model.predict() 打印张量 [[NaN],]

Question

我对机器学习和 tensorflow.js 完全陌生，我正在尝试预测下一组的值，但结果是“NaN”。 我究竟做错了什么 ？

按照这个 Github 示例

 async function myFirstTfjs(arr) {
    // Create a simple model.
    const model = tf.sequential();
    model.add(tf.layers.dense({units: 1, inputShape: [2]}));

    // Prepare the model for training: Specify the loss and the optimizer.
    model.compile({
      loss: 'meanSquaredError',
      optimizer: 'sgd'
    });
    const xs = tf.tensor([[1,6],
        [2,0],
        [3,1],
        [4,2],
        [5,3],
        [6,4],
        [7,5],
        [8,6],
        [9,0],
        [10,1],
        [11,2],
        [12,3],
        [13,4],
        [14,5],
        [15,6],
        [16,0],
        [17,1],
        [18,2],
        [19,3],
        [20,4],
        [21,5],
        [22,6],
        [23,0],
        [24,1],
        [25,2],
        [26,3]]);
    const ys = tf.tensor([104780,30280,21605,42415,32710,30385,35230,97795,31985,34570,35180,30095,36175,57300,104140,30735,28715,36035,34515,42355,38355,110080,26745,35315,40365,30655], [26, 1]);
    // Train the model using the data.
    await model.fit(xs, ys, {epochs: 500});
    // Use the model to do inference on a data point the model hasn't seen.
  model.predict(tf.tensor(arr, [1, 2])).print();
  }
  myFirstTfjs([28,5]);

Answer 1

正在发生的事情是ys中的大值导致非常大的错误。 这个大错误与（默认）学习率相结合，导致模型过度校正并且不稳定。 如果降低学习率，模型将收敛。

const learningRate = 0.0001;
const optimizer = tf.train.sgd(learningRate);

model.compile({
  loss: 'meanSquaredError',
  optimizer: optimizer,      
});

Answer 2

尝试将您的输出转换为更具可读性并更改您的优化器

var pred = model.predict(tf.tensor(arr, [1, 2]));
    var readable_output = pred.dataSync();
    console.log(readable_output);